Neural Correlates of Emotional Memory Biases in Adult Attachment: An Event‐Related Potential Study

ABSTRACT

Attachment‐related defences are theorized to systematically distort emotional information processing, yet the temporal dynamics of these biases remain poorly understood. The present study elucidated the neurophysiological mechanisms underlying implicit and explicit memory biases in attachment anxiety and avoidance using event‐related potentials (ERPs). University students completed both implicit and explicit memory tasks involving attachment‐relevant stimuli (threat and positive). Behavioural data revealed a double dissociation; the anxious group exhibited heightened memory sensitivity to threat, whereas the avoidant group showed memory bias toward positive stimuli. ERP findings indicated that avoidant individuals exhibited elevated P200 amplitudes early in processing, followed by a significant inhibition of the late parietal component (LPC). Conversely, attachment anxiety was characterized by elevated FN400 and sustained LPC amplitudes across all emotional stimuli. These results indicate that attachment dimensions differentially modulate distinct stages of neural processing, suggesting that defensive strategies are activated at specific temporal points to either amplify or inhibit specific emotional information.

Attachment anxiety and avoidance differentially bias emotional memory and distinct temporal stages of neural processing. Anxiety is characterized by enhanced threat memory and sustained late cortical activity, whereas avoidance shows early attentional engagement followed by inhibition of late emotional elaboration.

Abbreviations

ECR‐R

Experience in Close Relationships Scale‐Revised Questionnaire

ERP

Event‐related potential

LMM

linear mixed‐effects models

LPC

late parietal component

LPP

late parietal positivity

1. Introduction

1.1. Attachment and Regulatory Strategies

Early experiences with attachment figures directly affect brain development and significantly influence cognitive functions. It has long been assumed that the internal working models and mental representations of individuals and their relationship partners are based on their experiences in close relationships (Bowlby 1982). When attachment figures are unavailable or unreliable, a negative internal working model forms, leading to an insecure attachment style that can manifest as attachment anxiety or attachment avoidance (Fraley et al. 2006; Mikulincer and Shaver 2007).

This internal working model is the basis of developing strategies to regulate and enhance the monitoring of attachment‐related thoughts. Attachment theorists (Cassidy and Kobak 1988; Main 1990) have defined two secondary attachment strategies for processing attachment‐related information: hyperactivation and deactivation. People with an anxious attachment style develop hyperactivating strategies, tending to remain chronically alert to threats to self and others (Mikulincer et al. 2002; Mikulincer and Shaver 2010). Conversely, attachment‐avoidant individuals develop deactivating strategies, characterized by minimizing proximity‐seeking and actively suppressing attachment‐related thoughts (Edelstein and Gillath 2008; Mikulincer and Shaver 2003). Critically, recent perspectives suggest that deactivation is not a passive lack of emotion, but an active, effortful regulatory process that requires significant cognitive and physiological resources (Dozier and Kobak 1992). In comparison, hyperactivation consumes cognitive resources due to sustained attention and heightened vigilance. This illustrates the functional trade‐offs, where hyperactivation demands ongoing mental energy, while deactivation draws on cognitive reserves for active suppression.

1.2. Attachment and Information Processing

When processing attachment‐related information, individuals evaluate cues that may have detrimental effects on their well‐being. According to Mikulincer and Shaver (2003), these evaluations often occur at preconscious levels. In anxious attachment, hyperactivation appears as hypervigilant attention to psychological discomfort. Anxiously attached individuals are unable or unwilling to suppress access to threat‐related feelings, leading to mental rumination that may paradoxically result in inaccurate memory due to noisy or poorly encoded representations.

In contrast, avoidant individuals utilize deactivating strategies, which are divided into two categories: preemptive and postemptive strategies (Fraley and Shaver 1997). Preemptive strategies involve the early avoidance of unwanted cues, successfully decreasing the quantity of information encoded. Postemptive strategies involve the active suppression of memories that have already been encoded (Fraley, Garner, and Shaver 2000; Edelstein 2006a, 2006b). Research has shown that highly avoidant individuals are skilled at inhibiting attention to both positive and negative attachment‐related words (Edelstein and Gillath 2008) and show selective attention deficits in both threatening and positive contexts (Dewitte et al. 2007).

Evidence regarding memory biases in attachment is still inconsistent. Avoidant attachment is consistently linked to poorer memory for attachment‐related stimuli (Edelstein 2006a; Fraley and Brumbaugh 2007). However, findings for attachment anxiety are more controversial; some studies show greater memory performance (Gillath et al. 2005), while others find no significant relationship (Edelstein et al. 2005). These discrepancies highlight the need for a neurophysiological approach that can track the real‐time processing of attachment information. Specifically, investigating the dissociation between implicit and explicit memory systems is essential for clarifying whether attachment‐related biases originate from an initial failure in automatic encoding or from a subsequent strategic inhibition of conscious recollection—a distinction that purely behavioural metrics cannot resolve.

1.3. Neurophysiological Basis of Attachment‐Related Memory

To investigate the neurocognitive mechanisms of attachment‐related memory biases, it is important to distinguish between implicit and explicit processes. Implicit memory involves the automatic, unconscious influence of past experiences, while explicit memory requires the conscious, effortful retrieval of information. Here, implicit memory is operationally defined as memory that does not require conscious recollection and is often reflected in changes in behaviour or task performance due to prior experiences that are not explicitly remembered. Explicit memory, conversely, involves intentional recall of factual information, events or specific stimuli and is typically measured by tasks requiring individuals to deliberately retrieve stored information (Schacter 1987; Squire 2004). The main goal of this study is to investigate the implicit and explicit components of recognition memory in individuals with anxious and avoidant attachment styles. To this end, we focus on three specific event‐related potential (ERP) components: the P200, FN400 and late parietal component (LPC) amplitude to provide a temporal window into these distinct processes as they unfold during the recognition phase.

The P200 (typically 150–250 ms), often distributed over anterior regions, has been linked to early attentional allocation, stimulus orienting, task‐related salience (Carretié et al. 2001; Olofsson et al. 2008) and it is an index of early selective attention, reflecting the combination of top–down expectations and bottom–up sensory input (Stuellein et al. 2016). In research on emotions, the P200 is sensitive to stimulus salience (Carretié et al. 2004). For avoidant individuals, the P200 may serve as an early marker of hyper‐vigilance toward threat. Contrary to the view that avoidant individuals ignore threats, the physiological cost model suggests they may show increased early neural sensitivity (an early alarm system), allowing them to detect cues quickly enough to initiate suppression (Dozier and Kobak 1992; Zhang et al. 2008).

The FN400 (300–500 ms), typically maximal over mid‐frontal electrode sites, is the neurophysiological correlate of familiarity, an automatic form of knowing without contextual detail (Rugg and Curran 2007; Yonelinas 2002). In attachment research, the FN400 reflects the degree to which an attachment cue matches internal working models. For instance, an attachment cue (e.g., a partner's gesture or tone of voice) may evoke a sense of recognition based on past relational experiences. Anxious individuals, characterized by chronically accessible attachment schemas, often exhibit enhanced FN400 old/new effects (reduced negativity for previously seen stimuli), as their brains signal a high degree of emotional and semantic familiarity with attachment cues (Mikulincer and Shaver 2007; Zhai et al. 2016). Such cues create a strong, automatic sense of familiarity because they resonate with entrenched expectations formed through early and ongoing relational interactions. Conversely, Irak et al. (2020) found that avoidant individuals exhibit reduced FN400 amplitudes for emotional faces, suggesting that early encoding biases lead to the postemptive inhibition of emotional retrieval.

We focused on the LPC, a centro‐parietally distributed positive‐going wave occurring between 400 and 800 ms, which serves as a neurophysiological marker of conscious recollection and retrieval of contextual details (Rugg and Curran 2007). In the context of emotional stimuli, this component is often referred to as the late positive potential (LPP), reflecting sustained attentional engagement and the elaborative processing of emotional content (Schupp et al. 2006; Olofsson et al. 2008). The LPC component is particularly sensitive to postemptive defences. In avoidant individuals, a temporal dissociation may occur: while early and mid‐latency stages (P200, FN400) remain intact or even heightened—reflecting an initial sensitivity to attachment cues—the LPC is strategically inhibited to prevent the stimulus from entering conscious, elaborative processing (e.g., Domic‐Siede et al. 2025; Edelstein 2006b; Yan et al. 2022; Zheng et al. 2015). This late‐stage inhibition represents the functional neural marker of deactivating strategies, especially inside a self‐referential encoding framework. Research by Chavis and Kisley (2012) suggests that attachment anxiety is characterized by an attentional bias toward negative content, as evidenced by significantly larger differences in LPC amplitudes when viewing negative images compared with neutral ones.

1.4. The Current Study

This study investigates the influence of secondary attachment strategies on implicit and explicit memory for attachment‐related information. Guided by this neurophysiological framework, we achieve this by comparing ERPs recorded during the recognition phase across two encoding conditions: incidental encoding (Implicit Memory Task: self‐referent imagination) and intentional encoding (Explicit Memory Task), using attachment‐related threat and positive stimuli.

While hyperactivation involves constant monitoring for threat (Mikulincer and Shaver 2003), deactivation may also require significant early cognitive resources to detect and subsequently suppress such information (Dozier and Kobak 1992; Dewitte et al. 2007). Since P200 reflects early attentional allocation to semantically relevant emotional cues, we expect both groups to show heightened neural sensitivity to them. Our first hypothesis is that avoidant individuals will exhibit larger P200 amplitudes overall. This suggests a high level of initial physiological arousal or vigilance for suppression. Furthermore, because explicit memory tasks require a deliberate, top–down allocation of attention to stimuli (e.g., Hajcak et al. 2010), we hypothesize that the demands of explicit memory will enhance group‐specific early attentional biases. Specifically, in line with hyperactivation strategies (Mikulincer and Shaver 2003), we expect anxious individuals to exhibit increased P200 amplitudes to threat stimuli. Conversely, during the explicit memory task, we anticipate that avoidant individuals will show a selective reduction in the P200 response to old‐threat stimuli relative to positive stimuli. This early redirection of attention toward safe cues and away from threat demonstrates the earliest stage of preemptive defence (Fraley and Shaver 1997).

Our second hypothesis concerns the mid‐latency familiarity stage (Curran 2000). In the explicit memory task, we expect both groups to show a strong Old/New effect, with Old stimuli eliciting greater negativity—a pattern often associated with the processing of emotionally salient or conflicting information (Kutas and Federmeier 2011). However, we predict that anxious individuals will show significantly larger FN400 amplitudes than avoidant individuals across tasks. This suggests that the hyperactivation strategy (Mikulincer and Shaver 2007) increases the familiarity signal, while avoidant individuals already begin to dampen such signals as part of their regulatory process (Zheng et al. 2015). To counter the intentional retrieval demands of the explicit task, we expect avoidant individuals to actively suppress these early familiarity signals. Conversely, anxious individuals are likely to maintain heightened familiarity responses across both tasks due to consistent hyperactivation.

Our third hypothesis is that deactivation strategies will be most pronounced during the implicit memory task, where cognitive resources are not consciously directed toward retrieval (e.g., Edelstein 2006a, 2006b; Mikulincer et al. 2002). The implicit task relies on automatic memory retrieval processes, minimizing active engagement with the information and thereby reducing the influence of conscious control. This setup mirrors real‐world scenarios where avoidant individuals are prompted to rely more on ingrained, automatic processes rather than deliberate recollection, making it an ideal setting for observing deactivation strategies. We predict that avoidant individuals will exhibit lower LPC amplitudes, particularly for threat stimuli, while anxious individuals will maintain higher LPC amplitudes regardless of emotional valence. This would provide neurophysiological evidence for postemptive defence, where avoidant individuals successfully shut down the elaborative recollection process for distressing information (Fraley, Garner, and Shaver 2000). In addition, following previous literature suggesting a defensive exclusion (Bowlby 1980) of threat‐related details from conscious processing, we expect positive stimuli to elicit larger LPC amplitudes than threat stimuli. Generally, we expect the high cognitive load of the explicit memory task to limit the processing of emotional details. As a result, the neurophysiological difference between threat and positive stimuli observed in the implicit task should be reduced during intentional retrieval. Finally, we expect these neurophysiological patterns to be reflected in behavioural performance. Our fourth hypothesis is that anxious individuals will exhibit an increased propensity for false positives (higher false‐alarm rates) across encoding conditions, due to the amplified familiarity signals (FN400) and hypervigilance (Mikulincer and Shaver 2007). In contrast, avoidant individuals will exhibit diminished recognition accuracy (lower hit rates) for attachment‐related stimuli (Edelstein 2006a). This behavioural pattern in avoidant individuals is expected to stem from a clear dissociation reliance on between an intact FN400 (familiarity) while actively suppressing the LPC amplitude (recollection) to prevent conscious retrieval of distressing information.

2. Method

2.1. Participants

Forty‐four volunteer undergraduate students (28 female) between the ages of 18 and 25 years (M = 21.59, SD = 2.21) were tested. They were selected from a subject pool of 330 students who completed the Turkish version of the Experience in Close Relationships Scale‐Revised Questionnaire (ECR‐R; Selcuk et al. 2005), which measures attachment dimensions in the context of general close relationships (Fraley, Waller, and Brennan 2000). To maximize the contrast between hyperactivating and deactivating strategies and to ensure the categorical purity of the groups, we employed an extreme‐groups design, a methodology successfully utilized in previous attachment‐related ERP studies (e.g., Zheng et al. 2015). After screening, among these, participants who scored in both the highest quartile on the avoidance dimension of the ECR‐R (M = 82.83, SD = 15.89) and the lowest quartile on the anxiety dimension (M = 36.10, SD = 14.68) were assigned to the avoidant attachment group (N = 23, 13 female, Mage = 21.11, SDage = 1.49), and participants who scored in both the highest quartile on the anxiety dimension (M = 87.33, SD = 10.64) and the lowest quartile (M = 37.33, SD = 16.73) on the avoidance dimension were assigned to the anxious attachment group (N = 21, 15 female, Mage = 22, SDage = 2.65). This selection criterion ensured that we compared individuals with clear, non‐overlapping attachment profiles. Participants who reported severe vision problems, a history of neurological and/or psychiatric disorders, regularly used drugs of any sort that affect the central nervous system, or were left‐handed were excluded. Participants were screened using the short version of the MINI International Neuropsychiatric Interview (Sheehan et al. 1998), and only those without predisposition to psychopathological conditions were included. Participants with past and current psychiatric and/or neurological disorders or those taking medications that could affect cognitive processes were excluded.

2.2. Materials

2.2.1. Experiences in Close Relationships‐Revised (ECR‐R) Questionnaire

The Turkish version of the ECR‐R (Selcuk et al. 2005) developed by Fraley, Waller, and Brennan (2000) was used to assess participants' global romantic attachment orientations. The instructions directed participants to respond based on their general patterns in romantic relationships rather than on a specific current partner, and asked those not currently in a relationship to respond based on how they typically feel in such a context. Participants completed the 36‐item scale, which comprised two attachment dimensions: avoidance (18 items) and anxiety (18 items). They rated themselves on each item with a 7‐point Likert scale (‘strongly disagree’ to ‘strongly agree’). In scoring, 14 items were reverse‐scored (12 from the Avoidance subscale and two from the Anxiety subscale). The Cronbach's alpha coefficient for the avoidance subscale was 0.90, and for the anxiety subscale was 0.86. The test–retest reliability coefficients for the avoidance and anxiety dimensions were 0.81 and 0.82, respectively.

It should be noted that the ECR‐R (and its Turkish validation, YİYE‐II) conceptualizes attachment as a dimensional construct rather than a categorical one (Selçuk et al. 2005). Within this framework, security is defined as the lower end of both anxiety and avoidance dimensions, rather than as a distinct category. The primary objective of the present study was to compare the neurophysiological dynamics of two secondary attachment strategies (hyperactivation and deactivation), which are represented by high levels of anxiety and avoidance, respectively. To maximize the contrast between these divergent mechanisms, an extreme‐groups design was employed, focusing on individuals scoring in the highest quartiles of each dimension.

2.2.2. Stimuli

In the study, a total of 308 attachment‐related words, 154 threatening and 154 positive, were selected from previous studies (e.g., David 2009; Edelstein and Gillath 2008; Haydon et al. 2011; Izetelny 2006) and translated into Turkish to be used in implicit and explicit memory tasks. Three alternative Turkish translations were created for each word, and five experts conducting attachment research evaluated the translations, assigning each word to a category (threat or positive). The selection of stimuli was intentionally limited to attachment‐relevant emotional categories. According to attachment theory, secondary defensive strategies (hyperactivation and deactivation) are mainly triggered by emotionally salient, attachment‐relevant cues rather than neutral information. Therefore, attachment‐unrelated or neutral words were excluded, as they typically do not activate the attachment system. Interrater reliability analysis showed a Fleiss' Kappa value of κ = 0.79, indicating substantial agreement among the five raters (Landis and Koch 1977). Given the relatively simple dichotomy (threat vs. positive), this level of agreement suggests a high degree of consistency in the categorization process. Therefore, the words were divided into two categories based on their suitability. Then, for each experiment, two separate word lists were randomly generated, containing an equal number of words from the positive and threat categories.

2.2.3. Memory Tasks

The design of both experiments is the same except for two details. First, in the implicit memory task, participants were not told that they would be taking a memory test, and second, a filler task (Tower of Hanoi Task) was administered between the learning and recognition tests. In contrast, in the explicit memory task, participants were informed that it was a memory task, and no filler task was used. Additionally, a separate word list was used in each memory task.

The implicit memory task was performed in three stages (see Figure 1). In the first stage, 52 positives (e.g., friendship) and 52 threatening (e.g., unfaithful) attachment‐related words were used. Each word was presented for 10 s. The first and last two words of each list—consisting of two positive and two threat word per end—were excluded as buffer items to control for primacy and recency effects. Participants were informed that this is a self‐referent imagination task. When each word was displayed on the screen, participants were asked to imagine a scene or moment in which they were personally involved with the word. After imagining each scene or moment, they were asked to rate the degree of likes or dislikes for each imagined moment or scene on a 5‐point Likert scale (1 = not at all satisfied, 5 = completely satisfied). Before proceeding to the second stage, participants were given one trial of the Tower of Hanoi task as a filler task (maximum 5 min). In the second stage, participants were presented with the first letter of each word in a surprise word‐completion task to measure implicit memory. Participants were asked to write down the word presented in the first step, starting with the letters shown on the screen. In the final stage, in addition to old words, 50 new attachment‐related words (25 positive and 25 threat) were presented on the screen (2 s). The interval of each stimulus was 1 s. Participants were asked to indicate whether they were presented in the first phase for each word on the screen by pressing the left (yes) or right (no) mouse buttons. Except the buffer words, the order of the words was determined randomly.

FIGURE 1.

Schematic representation of experimental tasks: (a) Self‐referential encoding phase, where participants performed pleasantness judgements on imagined word‐list stimuli; (b) Word stem completion task, presented as an incidental test for the implicit memory condition; and (c) Recognition phase utilizing a standard old/new paradigm.

In the explicit memory task, during the encoding stage, participants were informed that their memory would be tested later. After the learning phase, participants performed a similar word‐completion task, and in the last phase, a recognition task was administered.

2.2.4. EEG Recording and Analysis

EEG responses were recorded during recognition phases. The decision to focus EEG analysis on the recognition (retrieval) stage was based on the incidental nature of the encoding task. During encoding, participants engaged in self‐referential imagination rather than intentional memorization; therefore, retrieval‐stage ERPs offered a more direct measure of how attachment‐related internal working models influence the accessibility and conscious processing of stored emotional information. EEG/EOG signals were recorded for 1200 ms after stimulus onset using 32 Ag/AgCl electrodes mounted in elastic Quick‐caps (Neuromedical Supplies, Compumedics Inc., Charlotte). EOG signals were measured from two bipolar channels: one was formed by two electrodes placed at the outer canthus of each eye and another by two electrodes below and above the left eye. EEG signals were recorded from 30 electrodes (FP1, FP2, F7, F8, F3, F4, Fz, FT7, FT8, FC3, FC4, FCz, T7, T8, C3, C4, Cz, TP7, TP8, CP3, CP4, CPz, P7, P8, P3, P4, Pz, O1, O2, Oz) arranged according to the standard 10–20 system, with additional electrodes placed at BP1/BP2 and also on the left and right mastoids (M1/M2). All EEG electrodes were referenced online to a vertex electrode and re‐referenced offline to linked mastoids. EEG/EOG signals were amplified and recorded at a 1000 Hz sampling rate using a Synamp2 amplifier in AC mode (NeuroScan, Compumedics Inc., Charlotte) with high‐ and low‐pass filters set at 0.15 and 100 Hz, respectively. EEG electrode impedance was kept below 5 kΩ. EEG data pre‐processing was conducted using Edit 4.5 (Neuroscan, Compumedics Inc., Charlotte) and applied to each participant's dataset. To ensure objectivity, different researchers conducted data collection and analysis. The analyst only received de‐identified data files with participant ID numbers, without any group assignment or experimental condition details; therefore, the analyst remained fully blinded to group membership during the statistical analysis. Data were down‐sampled to 250 Hz to reduce computational demands and then low‐pass filtered at 30 Hz and high‐pass filtered at 0.15 Hz. EEG segments were extracted with an interval of 100 ms preceding and 1000 ms following the prime stimuli onset. During the artefact rejection procedure, EEG recordings from each participant were first spatially filtered. These data were then filtered with a band‐pass filter centred at 0.15 and 31 Hz (12 Oct/dB). Noisy parts on continuous data were rejected with 100 microvolts of spatial filtering before epoching. Continuous data, whose markers were placed before averaging, were epoched from −100 to 1000 ms and used in the pre‐stim baseline procedure. Then, the averaging procedure was applied to the obtained epoch directory. Lastly, an expert rater performed manual visual inspection to remove any remaining trials contaminated by non‐stereotypical artefacts or saccadic movements, specifically noting step‐like changes in the horizontal EOG channels. The individual averages were turned into appropriate grand averages with the grouping process. The full analysis pipeline is part of the script code in Neuroscan 4.51 Batch Editor. For ERP computation, artefact‐free segments were baseline‐corrected using a 100‐ms prestimulus period and then averaged across the experimental condition. In the present study, the average ERP was determined along the temporal axis. To obtain the overall average, epochs were acquired and filtered from 44 participants, with epochs derived according to the task phase and/or response type. Accordingly, grand averages for correct recognition of old and new items were calculated for two groups (anxious and avoidant) and two emotion types (threat and positive) for each memory task separately. Thus, grand averages were calculated for 15 electrodes grouped into five regions of interest (ROIs): frontal (F3, Fz, F4), fronto‐central (FC3, FCz, FC4), central (C3, Cz, C4), centro‐parietal (CP3, CPz, CP4) and parietal (P3, Pz, P4). Since we did not have specific hypotheses about laterality, electrode waveforms were averaged within each region to create a single regional waveform for subsequent analysis. For example, electrodes F1, Fz and F2 were averaged into one ‘frontal’ waveform [(F1 + Fz + F2)/3]. Three ERP components were identified and analysed at regions where their activity was maximal (Eimer 2000; Luck 2014): the P200 (150–240 ms) was analysed at frontal, fronto‐central, central and centro‐parietal regions; the FN400 (300–500 ms) at frontal, fronto‐central and central regions; and the LPC (500–800 ms) at centro‐parietal and parietal regions. All participants who had at least 20 artefact‐free trials per condition were included in the analyses (two subjects were excluded).

2.3. Procedure

The host University's Research Ethics Committee approved the study, and written informed consent was obtained from all participants before their involvement. Following completion of ECR‐R, eligible participants (n = 97) were screened according to the EEG and study screening protocol. According to this protocol, 41 participants were excluded (e.g., due to current or previous medical history, left‐handedness or current medication use). Also, eight individuals did not respond to e‐mails or phone calls for appointments. Finally, 48 participants were recruited. However, four participants were excluded from the final sample, two due to equipment failure or excessive EEG artefacts, and two because they provided fewer than 20 artefact‐free trials per condition. Thus, statistical analyses were performed with 44 participants. Participants were advised to abstain from sleep deprivation, alcohol, and caffeine on the evening before the study. Participants were introduced to the laboratory setting and provided with the study information sheet. Orders of the experiments were counterbalanced. The full session lasted approximately 2 h per participant.

2.4. Sample Size Calculation

A post hoc sensitivity analysis was conducted using G*Power (Faul et al. 2007) to determine the minimum detectable effect size for our final sample of N = 44. Given an alpha of 0.05 and a standard power criterion of 0.80, our study was sufficiently powered to detect a medium effect size of approximately ƒ = 0.22 (ƞ _p ² = 0.04) for the interaction effects in our mixed‐design framework. This suggests that the study was well‐positioned to identify psychologically meaningful differences in ERP amplitudes. This sample size (N = 44) is comparable with or larger than those in several previous ERP studies on attachment‐related memory biases, which have produced strong results with groups of similar or smaller size (e.g., Zhang et al. 2008, N = 30; Zheng et al. 2015, N = 26). Furthermore, the use of LMM maximized statistical power by utilizing the full distribution of trial‐level data (with a minimum of 20 artefact‐free trials per condition per participant) rather than relying on a single aggregated mean per subject.

3. Results

We analysed ERP amplitudes using linear mixed‐effects models (LMMs) in IBM‐SPSS (version 23). The LMMs were selected instead of repeated‐measures ANOVA to address sphericity violations (Gueorguieva and Krystal 2004), account for spatial autocorrelation (Litvak et al. 2007) and accommodate the variable number of regions analysed per component. The models incorporated fixed effects for task, group, emotion, stimulus type, region (only for ERP data) and their interactions. Random intercepts for subjects (N = 44) were included to account for baseline differences (Baayen et al. 2008). Estimation utilized restricted maximum likelihood with Type III sums of squares. Analyses were conducted separately for each task and for three ERP components.

3.1. Behavioural Results

To examine how attachment style influences emotional memory processing across different retrieval tasks, a subsequent LMM was conducted on emotion‐specific d′ scores. The model included group (anxious vs. avoidant) as a between‐subjects factor, and task type (implicit vs. explicit) and emotion type (threat vs. positive) as within‐subjects factors.

The random intercept for participants was significant, Estimate = 0.196, SE = 0.054, Wald Z = 3.76, p < 0.001. The analysis showed no significant main effects for group, F(1, 40.79) = 0.45, p = 0.504; task type, F(1, 117.90) = 0.35, p = 0.555; or emotion type, F(1, 115.99) = 0.89, p = 0.347. The interaction between group and task type, F(1, 117.90) = 0.02, p = 0.891, and the three‐way interaction among group, task type and emotion type, F(1, 115.99) = 0.04, p = 0.844, were also not significant. These results suggest that baseline memory capacity and task requirements were well matched and that attachment‐related emotional biases were unaffected by whether the retrieval task was implicit or explicit.

Furthermore, a notable interaction effect between task and emotion type was observed, F(1, 115.99) = 17.88, p < 0.001, indicating general differences in emotion processing across tasks regardless of attachment group. Supporting our hypothesis, the LMM revealed a significant two‐way interaction between group and emotion type, F(1, 115.99) = 6.50, p = 0.002. To examine this interaction, post hoc pairwise comparisons with Bonferroni correction were performed. The results showed a double dissociation in which the anxious group exhibited significantly higher memory sensitivity to threat words (M = 0.28, SE = 0.12) than to positive words (M = 0.10, SE = 0.09), p = 0.006. Conversely, the avoidant group showed significantly greater memory sensitivity to positive words (M = 0.29, SE = 0.09) than to threat words (M = 0.11, SE = 0.11), p = 0.027. These results confirm that anxious and avoidant individuals employ different, attachment‐related defensive strategies when processing emotional information, whether the memory task is implicit or explicit.

3.2. Electrophysiological Results

Figures 2 and 3 display the ERP waveforms elicited during correct responses to threat and positive stimuli for old (A) and new (B) items across the attachment orientations for the implicit and explicit memory tasks, respectively. Similarly, Figures 4 and 5 present the corresponding topographical scalp maps for both stimulus types and attachment groups across memory conditions.

FIGURE 2.

Grand average ERP waveforms for the implicit memory task. Waveforms were recorded during correct responses to positive (solid line) and threat (dashed line) stimuli. Data are shown for old (A) and new (B) items across anxious (red) and avoidant (black) attachment orientations at 12 electrodes, grouped into four regions: frontal, fronto‐central, central, and parietal (from top to bottom row). The rightmost column displays the region‐averaged ERP. Stimulation occurred at the ‘0.0 ms’ time point (stimulus onset). Shaded areas represent the standard error of the mean (SEM).

FIGURE 3.

Grand average ERP waveforms for the explicit memory task. Waveforms were elicited during correct responses to positive (solid line) and threat (dashed line) stimuli. Data are presented for old (A) and new (B) items across anxious (red) and avoidant (black) attachment orientations at12 electrodes, grouped into four regions: frontal, fronto‐central, central, and parietal (from top to bottom row). The rightmost column shows the region‐averaged ERP. Stimulation applied at ‘0.0 ms’ time point (stimulus onset). Shaded areas indicate the standard error of the mean (SEM).

FIGURE 4.

Topographical scalp maps for correct responses during the implicit memory task. Maps are displayed for old (left) and new (right) items. The first and second rows represent the anxious and avoidant groups, respectively, with sub‐sections for positive (top) and threat (bottom) stimuli. Scalp distributions are presented across the three analysed ERP time windows.

FIGURE 5.

Topographical scalp maps for correct responses during the explicit memory task. Maps are displayed for old (left) and new (right) items. The first and second rows represent the anxious and avoidant groups, respectively, with sub‐sections for positive (top) and threat (bottom) stimuli. Scalp distributions are presented across the three analysed ERP time windows.

3.2.1. P200 Amplitude

The random intercept for participants was significant, Estimate = 0.388, SE = 0.092, Wald Z = 4.20, p < 0.001. The LMM analysis revealed a significant main effect of group on the P200 amplitude, F(1, 41.56) = 5.26, p = 0.027, indicating that the P200 amplitude was larger for the avoidant group (M = 1.62, SE = 0.14) compared with the anxious group (M = 1.09, SE = 0.15), p = 0.011. There was a significant main effect of emotion, F(1, 622.32) = 111.45, p < 0.001, showing that positive stimuli (M = 1.78, SE = 0.12) elicited significantly greater amplitudes than threat stimuli (M = 0.39, SE = 0.13), p = 0.009. The main effect of region was also significant, F(2, 622.32) = 24.12, p < 0.001. Follow‐up contrasts indicated that P200 amplitudes at the frontal (M = 1.99, SE = 0.17) and fronto‐central (M = 0.81, SE = 0.15) regions did not significantly differ from each other (p = 0.181), but were larger than those at the centro‐parietal (M = 0.67, SE = 0.15) and central (M = 0.66, SE = 0.14) regions (all ps < 0.001). Additionally, a significant main effect of stimulus type was observed, F(1, 622.32) = 16.59, p < 0.001. The old stimuli (M = 2.94, SE = 0.10) produced bigger P200 amplitudes than new stimuli (M = 1.69, SE = 0.11), p = 0.010.

The interaction effect between group and emotion type was significant, F(1, 622.32) = 11.23, p = 0.001. Follow‐up analyses showed that, for the avoidant group, the P200 amplitude was significantly smaller for threat stimuli (M = 1.84, SE = 0.15) than for positive stimuli (M = 2.78, SE = 0.14), p = 0.001. Conversely, for the anxious group, threat stimuli (M = 2.88, SE = 0.11) generated a significantly larger P200 amplitude than positive stimuli (M = 1.79, SE = 0.16), p = 0.001.

Finally, the LMM analysis revealed a significant three‐way interaction between group, emotion type, and task type, F(1, 481.18) = 9.37, p = 0.007. In the anxious group, the P200 amplitude to threat stimuli was stronger during the explicit task (M = 2.82, SE = 0.16) than to positive stimuli (M = 1.48, SE = 0.16), p = 0.001. While this threat bias persisted in the implicit task, the amplitudes were notably reduced (MThreat = 1.60, SE = 0.15; MPositive = 1.04, SE = 0.18). Conversely, for the avoidant group, the increased P200 amplitude for positive stimuli was highest during the explicit task (M = 2.87, SE = 0.16) compared to threat stimuli (M = 1.24, SE = 0.15), p = 0.001. In the implicit task, this positive bias was also observed but considerably diminished (M_Positive = 1.71, SE = 0.17; M_Threat = 1.11, SE = 0.15), p = 0.009.

3.2.2. FN400 Amplitude

The random intercept for participants was significant, Estimate = 1.238, SE = 0.287, Wald Z = 4.31, p < 0.001. The LMM analysis revealed a significant main effect of group, F(1, 42.84) = 5.78, p = 0.038, indicating that the anxious group exhibited a significantly larger FN400 amplitude (M = −1.39, SE = 0.29) compared with the avoidant group (M = −0.89, SE = 0.40), p = 0.021. A significant main effect of stimulus type was also observed, F(1, 732.09) = 13.21, p < 0.001, reflecting a classic old/new effect where old stimuli (M = −0.57, SE = 0.01) elicited more negative amplitudes than new stimuli (M = −0.22, SE = 0.05), p = 0.001. Furthermore, a significant main effect was found for emotion type, F(1, 732.09) = 4.49, p = 0.030, with positive stimuli (M = −1.97, SE = 0.13) eliciting larger FN400 amplitudes than threat stimuli (M = −0.82, SE = 0.14), p = 0.001. The main effect of task type was also significant, F(1, 732.09) = 15.14, p < 0.001, indicating that the explicit memory task yielded a significantly larger FN400 amplitude (M = −1.81, SE = 0.17) than the implicit memory task (M = −0.47, SE = 0.16), p = 0.001.

A significant two‐way interaction effect between group and stimulus type was found, F(1, 732.09) = 8.77, p = 0.003. Post hoc pairwise comparisons revealed that the anxious group showed significantly larger negative amplitudes to old stimuli (M = −1.030, SE = 0.25) compared to new stimuli (M = −0.425, SE = 0.26, p = 0.001). Conversely, the avoidant group displayed no significant difference between old (M = −0.081, SE = 0.24) and new stimuli (M = −0.020, SE = 0.25), p = 0.411. Additionally, a significant interaction between stimulus type and task type was observed, F(1, 732.09) = 4.73, p = 0.030, indicating that the FN400 amplitude for old stimuli was grater during the explicit memory task (M = −1.41, SE = 0.18) compared with the implicit memory task (M = −0.35, SE = 0.17), p = 0.004, while responses to new stimuli across the task was not significant (p = 0.101).

Finally, the interaction effect between group, stimulus type and task type was significant, F(1, 732.09) = 4.00, p = 0.046. Post hoc analysis of this interaction showed that, in the explicit task, the anxious group displayed greater negativity to old stimuli (M = −1.22, SE = 0.11) compared with new stimuli (M = −0.36, SE = 0.10), p = 0.001. In contrast, during the explicit memory task, the avoidant group demonstrated a noticeably reduced FN400 response, with a small difference between old (M = −0.36, SE = 0.09) and new stimuli (M = −0.17, SE = 0.02), p = 0.036.

3.2.3. LPC Amplitude

The random intercept for participants was significant, Estimate = 0.970, SE = 0.228, Wald Z = 4.26, p < 0.001. The LMM analysis revealed a significant main effect of stimulus type, F(1, 770.38) = 4.72, p = 0.031, indicating that the old stimuli elicited significantly larger LPC amplitudes (M = 1.315, SE = 0.16) compared with new stimuli (M = 0.129, SE = 0.16), p = 0.001. Significant main effects were also found for emotion type, F(1, 770.38) = 17.22, p < 0.001, with positive words (M = 1.08, SE = 0.16) yielding higher amplitudes than threat words (M = 0.63, SE = 0.16), p = 0.047. The main effect of task type was significant, F(1, 770.38) = 44.12, p < 0.001, indicating larger amplitudes during the explicit memory task (M = 1.11, SE = 0.08) compared with the implicit memory task (M = 0.51, SE = 0.08), p = 0.001. Furthermore, a main effect of region was observed, F(1, 469.54) = 16.16, p < 0.001, with the centro‐parietal region showing higher amplitude (M = 1.14, SE = 0.16) compared with the parietal region (M = 0.49, SE = 0.18), p = 0.005.

A significant interaction between group and emotion type was observed, F(1, 770.38) = 5.97, p = 0.015. Post hoc comparisons revealed that the avoidant group showed significantly larger LPC amplitude to positive stimuli (M = 1.15, SE = 0.22) than to threat stimuli (M = 0.471, SE = 0.22), p = 0.001. In contrast, for the anxious group, LPC amplitudes between positive (M = 0.93, SE = 0.23) and threat stimuli (M = 0.88, SE = 0.23) were not significant (p = 0.689).

Finally, a significant interaction between emotion type and task type was found, F(1, 770.38) = 5.64, p = 0.018. During the implicit memory task, the LPC amplitude for threat words (M = 0.21, SE = 0.11) was significantly lower than for positive words (M = 0.81, SE = 0.11), p = 0.007. Conversely, during the explicit memory task, the difference in LPC amplitudes for positive (M = 1.195, SE = 0.11) and threat words (M = 1.025, SE = 0.11) was non‐significant, p = 0.817.

4. Discussion

The present study aimed to explain the neurophysiological mechanisms underlying emotional implicit and explicit memory biases in attachment anxiety and avoidance. By comparing implicit and explicit memory tasks, we identified a temporal dissociation in how secondary attachment strategies regulate information processing. Our results suggest that while both strategies are activated early, they differ markedly during the transition from automatic familiarity to conscious recollection and are influenced by the cognitive demands of the retrieval task.

4.1. Behavioural Findings

An analysis of behavioural variables, measured via sensitivity scores as well as hit rates and false alarms, revealed a distinct pattern of emotional memory across both tasks for the two groups. Specifically, the anxious group's hypervigilance was evident not only in their elevated hit rates for threat stimuli but also in their significantly increased false alarm rates for novel threat words, indicating a hyperactivating bias that linked to memory illusions. In the implicit memory task, a double dissociation in sensitivity was also observed. The anxious group showed greater sensitivity to threat stimuli than the avoidant group, whereas the avoidant group showed greater sensitivity to positive stimuli than the anxious group. These differences were reflected in the emotional bias index, which showed greater sensitivity to threat in the anxious group than in the avoidant group. In the explicit memory task, the patterns of sensitivity remained largely consistent with those found in the implicit task. The anxious group again showed greater sensitivity to threat stimuli than the avoidant group. While the avoidant group showed greater sensitivity to positive stimuli than the anxious group, this difference was less pronounced than that observed for threat stimuli. Despite this, the overall emotional bias toward threat remained higher in the anxious group than in the avoidant group.

The behavioural results of the current study provide robust evidence for the Hyperactivation–Deactivation Model of attachment (Mikulincer and Shaver 2003). By utilizing signal detection theory, we were able to isolate memory sensitivity from response bias, showing a clear double dissociation in how attachment anxiety and avoidance shape the encoding and retrieval of emotional information. Alternative models, such as the Monotropic Model and Emotion Dysregulation Model, were considered but less accurately explained the observed patterns. The Monotropic Model (Bowlby, 1988), which emphasizes a singular primary attachment figure, falls short in accounting for the dual strategies of anxiety‐dominated and avoidance‐dominated patterns observed in our data. Similarly, the Emotion Dysregulation Model (Mikulincer et al. 2003) fails to address the strategic regulation of emotional bias seen in this study. As hypothesized, anxious individuals demonstrated significantly greater memory sensitivity to threat stimuli than avoidant individuals, particularly in the implicit memory task. This finding is consistent with the hyperactivation strategy, characterized by a chronic accessibility of attachment‐related distress (Mikulincer and Shaver 2007).

According to the hypervigilance hypothesis, the anxious attachment system is programmed to detect and prioritize signs of rejection or abandonment. Our results suggest that this bias is not simply a conscious preoccupation but an automatic cognitive filter. The higher memory performance for threat in the anxious group indicates that these stimuli receive deeper, more elaborate processing, rendering them extremely resistant to forgetting. This all‐access pass for threat‐related information ensures that distress‐related cues continue to dominate the anxious individual's cognitive landscape.

In contrast, avoidant individuals exhibited a behavioural pattern consistent with deactivating strategies. Although they showed the lowest sensitivity to threat stimuli, they demonstrated significantly better memory for positive stimuli than the anxious group. This behavioural profile supports the concept of defensive exclusion (Bowlby 1980). To maintain emotional distance and avoid activating the attachment system, avoidant individuals appear to divert cognitive resources away from distress‐related cues strategically. However, it is important to note that some previous studies have failed to find such lower memory performance in avoidant individuals when the stimuli are not sufficiently attachment‐relevant (e.g., Edelstein 2006b) or when individuals are instructed to focus on the information for a secondary task (Fraley, Garner, and Shaver 2000). Our findings suggest that when stimuli are processed in a self‐referential context, defensive exclusion is particularly robust. This self‐referential processing appears to exacerbate the reduced memory performance observed in avoidant individuals, creating a clear contrast with non‐self‐referential tasks where such memory exclusion is less pronounced. Highlighting this difference illustrates the boundary conditions under which avoidant memory patterns emerge, underscoring the context‐dependent nature of these defensive strategies.

The higher memory for positive information, termed here as a positive distraction strategy, suggests that avoidant individuals may prioritize non‐threatening, positive cues as cognitive distractors. By focusing on positive stimuli, they effectively crowd out potential sources of distress, thereby maintaining their goal of affective homeostasis (Edelstein 2006b). A critical contribution of this study is the emergence of these biases within the implicit memory task. Because the incidental encoding phase (self‐referent imagination) did not require intentional memorization, the resulting differences represent the automatic operation of internal working models (Kirsh and Cassidy 1997). In this context, it is important to clarify that the emotional biases observed reflect the activation of attachment‐specific cognitive schemas rather than generic emotional experiences. Because the encoding task required participants to imagine themselves within the scenario, the stimuli directly engaged their relational schemas, rather than merely inducing transient positive or negative affect. The fact that avoidant individuals failed to recognize threat words even when they were processed in a self‐referential context suggests that their defensive filters are active from the very earliest stages of information intake. This outcome refines the view of avoidant defenses as merely conscious forgetting, suggesting instead a basic reconfiguration of neural storage priorities. Future research employing paradigms that specifically dissociate automatic encoding from controlled retrieval effort will be essential to confirm whether this reconfiguration occurs independently of conscious intent.

4.2. Electrophysiological Findings

An analysis of the P200 component, an ERP index of early sensory processing and attentional resource allocation, showed a significant main effect of group, with larger amplitudes in the avoidant group than in the anxious group. In the explicit memory task, a significant interaction between group and emotion indicated that the avoidant group showed larger amplitudes to positive stimuli than to threat stimuli, whereas the anxious group showed larger amplitudes to threat stimuli than to positive stimuli. Regarding the FN400 component, associated with automatic familiarity processing, a significant main effect of group was found, with the anxious group exhibiting larger amplitudes than the avoidant group across conditions. For the LPC, linked to conscious recollection and retrieval of episodic details, a significant interaction between group and emotion type emerged. The anxious group maintained high LPC amplitudes for both positive and threat stimuli, whereas the avoidant group selectively reduced LPC amplitudes specifically for threat stimuli compared to positive stimuli. Furthermore, a significant interaction between emotion type and task type was observed, showing that the processing of threat words was particularly diminished during the implicit memory task.

The P200 component serves as an index of early sensory gating and the allocation of attentional resources. A primary finding of this study was a significant main effect of group: avoidant individuals exhibited larger P200 amplitudes than anxious individuals. This data presents neurophysiological evidence for the regulatory cost of deactivation (Edelstein 2006b). Avoidant individuals do not ignore attachment indicators through passive indifference; rather, their brains engage in a resource‐demanding effort to monitor and filter incoming information. These findings, which contrast with some previous ERP research, suggest that early attentional biases for threat are primarily a hallmark of attachment anxiety (e.g., Zhang et al. 2008; Zheng et al. 2015). While some studies have reported no group differences in early components like the P200, our results suggest that the preemptive nature of avoidant defences may manifest as heightened early‐stage monitoring when stimuli are processed through a self‐referent lens.

The interaction effects between group and emotion in the explicit memory task further clarify this mechanism. For avoidant individuals, the significantly lower P200 amplitude to threat relative to positive stimuli shows a preemptive defence. This suggests that within 200 ms of stimulus onset, the avoidant brain attempts to down‐regulate the salience of distress‐related cues. Conversely, the anxious group's heightened P200 to threat relative to positive stimuli confirms a state of hypervigilance, in which the early orienting response is biased toward rapid detection of possible rejection (Mikulincer and Shaver 2007).

The FN400 is typically associated with familiarity (Curran 2000). Our results revealed that anxious individuals produced significantly larger amplitudes than the avoidant group. This finding provides neurophysiological support for the chronic accessibility hypothesis (Mikulincer 1998). According to this conceptual framework, hyperactivating strategies involve constant monitoring of the environment for cues of rejection, keeping attachment‐related schemas in a state of continuous readiness (Mikulincer and Shaver 2007). Our results suggest that, for anxious individuals, attachment‐related cues elicit a high degree of neural familiarity across task demands. This spontaneous schematic activation in response to attachment stimuli (Baldwin et al. 1996) suggests that hyperactivating strategies maintain attachment schemas in a state of constant readiness, making it difficult for these individuals to ignore even nuanced signals of distress.

The LPC component is typically associated with elaborative, conscious recollection and the retrieval of episodic details (Rugg and Curran 2007). The memory retrieval problem hypothesis proposed by some researchers suggests that avoidant individuals encode information successfully but fail to retrieve it (e.g., Fraley, Garner, and Shaver 2000; Fraley and Brumbaugh 2004). Instead, our data support the postemptive defence model (Fraley and Brumbaugh 2004; Edelstein 2006a, 2006b). Instead of a generalized low memory retrieval, our data support the postemptive defence model (Fraley and Brumbaugh 2004; Edelstein 2006b) through an emotion‐specific mechanism. The significant group‐by‐emotion type interaction in the LPC window shows that avoidant individuals do not completely shut down all elaborative processing; instead, they selectively reduce processing of threat. While the anxious group displayed heightened LPC amplitudes for all emotional stimuli, the avoidant group showed a significant decrease in LPC amplitude specifically in response to threat compared to positive stimuli. Our results suggest that even when attachment‐related stimuli are detected (P200) and categorized as familiar (FN400), deactivating strategies successfully inhibit the elaborative processing required for conscious recollection of threat. This apparent discrepancy between increased early neural responses (P200) and the subsequent absence of behavioural recognition highlights the core mechanism of deactivation strategies. This mechanism is consistent with Bowlby's (1980) concept of defensive exclusion, wherein distressing information is systematically blocked from conscious awareness immediately after initial detection to maintain affective homeostasis (Mikulincer and Shaver 2007). In this sense, the discrepancy between brain and behaviour is not a functional failure, but rather a hallmark of avoidant defence, an early‐stage monitoring of threat followed by a late‐stage ‘cognitive cutoff’ that prevents disconcerting information from reaching conscious episodic memory. This suggests that even when avoidant individuals are forced to acknowledge the presence of a threat, they strategically constrict the depth of neural processing. By selectively limiting the elaborative stage of memory specifically for threat cues, they prevent the stimulus from evolving into a vivid, emotionally distressing experience, thereby preserving their goal of cognitive and emotional self‐reliance (Mikulincer and Shaver 2007).

To synthesize findings across different cognitive systems, it is essential to examine the similarities and differences between implicit and explicit memory biases. For anxiously attached individuals, a high degree of similarity was observed across both systems; their hyperactivating strategies lead to a consistent threat bias, regardless of whether the task involves automatic processing or conscious recollection. Conversely, for avoidant individuals, a clear distinction exists between the two memory types. While their implicit system initially detects and processes emotional salience (as indicated by enhanced P200), they later demonstrate a strategic ‘shutting down’ of threat‐related information during conscious retrieval phases (as indicated by suppressed LPC for threat). The interaction between emotion and task type further highlights that this emotional modulation of late‐stage processing is adaptable and responds to explicit retrieval demands.

4.3. Conclusion, Limitations and Future Directions

In conclusion, we propose that attachment‐related defences operate through a multi‐stage regulatory architecture that filters information across the information processing sequence (Mikulincer and Shaver 2003, 2007). Our results suggest a fundamental divergence in how individuals control the flow of emotional information, transitioning from early sensory gating to late‐stage cognitive consolidation. Avoidant individuals appear to use a bimodal defensive strategy comprising both preemptive and postemptive mechanisms (Fraley and Brumbaugh 2004). This process begins with an initial down‐regulation of stimulus salience at the P200 stage, followed by an emotion‐specific inhibition of elaborative encoding for threat at the LPC stage. By effectively truncating the cognitive processing stream before it reaches the stage of conscious, episodic appraisal, the avoidant individuals ensure that distressing information remains fragmented. This temporal dissociation provides a robust neurobiological basis for the defensive exclusion and subjective memory deficits long associated with avoidant deactivation (Bowlby 1980; Edelstein 2006a).

In contrast, attachment anxiety is characterized by a feed‐forward hyperactivating cascade, and in this group, heightened attentional orienting (P200) and elevated schematic accessibility (FN400) act as catalysts for potentiated conscious recollection (LPC). Rather than filtering information, the anxious processing system appears to amplify it, ensuring that attachment‐related cues—especially those connected to threat—are prioritized for deep, ruminative processing (Mikulincer and Shaver 2007). This is consistent with the social‐cognitive view that anxious internal working models function as chronically accessible schemas that lower the threshold for threat detection (Baldwin et al. 1996). Ultimately, this study demonstrates that attachment styles do not merely influence what we remember but fundamentally dictate the temporal windows in which emotional information is allowed to persist. This data illustrates the refined quality of internal working models (Bowlby 1973), which operate as dynamic, real‐time filters that shape the neurocognitive architecture of our social world.

This study has limitations. First, the lack of a secure control group limits our ability to determine whether these neural signatures are unique to insecure attachment or represent extremes of a universal regulatory process. Although our extreme‐groups design specifically aimed to compare hyperactivating and deactivating strategies based on the dimensional model of the ECR‐R (Selcuk et al. 2005) rather than a categorical baseline, future studies using a full‐spectrum approach could establish a normative baseline, yielding deeper theoretical insights into these variations. Second, while reliably categorized word lists ensured high experimental control and construct validity, the absence of attachment‐unrelated neutral stimuli limits our ability to compare these biases against a non‐emotional baseline. Future research should replicate these findings using more complex stimuli, such as facial expressions of rejection or dynamic social scripts, to enhance translational relevance and ecological validity. Third, our university‐based sample restricts the generalizability of the findings to broader demographic or clinical populations and necessitates longitudinal studies to determine if the observed neural markers are malleable. Finally, given the limited spatial resolution of ERPs, further research using fMRI is needed to identify the specific neural substrates, such as prefrontal‐amygdala circuitry, underlying these defensive processes at different stages of memory. Such an approach would provide greater anatomical precision to the temporal findings reported here and offer a more holistic understanding of attachment‐related neurocognitive mechanisms.

Author Contributions

Metehan Irak: conceptualization, investigation, writing – original draft, methodology, validation, writing – review and editing, project administration, supervision. Öznur Çamoğlu: visualization, software, formal analysis, data curation.

Funding

No funding was received for conducting this study.

Ethics Statement

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Scientific Publication and Research Ethics Committee of Bahçeşehir University (Date‐No: 11/O4/2014‐014).

Consent

Informed consent was obtained from all individual participants included in the study.

Conflicts of Interest

The authors declare no conflicts of interest.

Irak, M. , and Çamoğlu Ö.. 2026. “Neural Correlates of Emotional Memory Biases in Adult Attachment: An Event‐Related Potential Study.” European Journal of Neuroscience 64, no. 2: e70621. 10.1111/ejn.70621.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.