Attraction in every sense: How looks, voice, movement and scent draw us to future lovers and friends

Abstract

What makes someone attractive has been examined for faces, in terms of common preferences, and for opposite‐sex interactions. We expanded on this by considering also other non‐verbal modalities, personal preferences and same‐sex interactions. We presented the face, body motion, voice and body odour from 61 non‐verbal agents (34 women) as stimuli in an attractiveness rating to 71 perceivers (37 women). Our results showed that the modalities were differently attractive and that some correlated more than others. Specifically, body odours were least and audio–video stimuli most attractive. Voice/looks as well as body odour/movement showed fairly robust positive associations. Both common and personal preferences accounted for variance in the data. Most effects compared between opposite‐ and same‐sex ratings, with only a few exceptions, including that only same‐sex ratings showed a clear dominance of personal over common preferences. We conclude that the different non‐verbal modalities are equally relevant for attraction but differ in absolute attractiveness and redundancy, likely due to their different suitability for communicating stable (e.g. genetic) versus variable (e.g. hormonal) person characteristics. Beauty excites agreement and disagreement; it matters not only in encounters with the other sex but in social interactions more broadly.

How we are liked by others is often decided within only a few hundred milliseconds (Willis & Todorov, 2006). Our physical appearance prompts split‐second assessments of attractiveness, which in turn shape first impressions on a range of personal characteristics. Research to date has pursued attractiveness primarily within the context of face perception and identified a few common characteristics (e.g. averageness) with relevance for mate selection. However, impression formation is typically a multisensory process that integrates a range of non‐verbal channels including, for example, vocal, motion and olfactory information (Schirmer et al., 2025). Moreover, its processes appear to rely not only on common but also personal preferences and have implications beyond mate selection. Thus, we wished to extend previous research by taking a more holistic perspective and by pursuing the role of interpersonal attraction in human relationships more broadly.

Who do we find attractive?

In 1966, Walster et al. (1966) discovered that physical attractiveness, but none of the other ratings they had recorded (e.g. intelligence) predicted whether college students wished to date a new, opposite‐sex acquaintance. This discovery ignited interest in attractiveness and led to the development of the field of attractiveness research. Since then, a growing number of studies have indicated that at least three factors make a person attractive: (i) averageness of their physical features, (ii) the extent to which these features are sex‐typical and (iii) how symmetrical they are (Gangestad & Scheyd, 2005; Little et al., 2011; Thornhill & Gangestad, 1993).

In this context, averageness refers to an appearance that is prototypical of men and women. Its role for attractiveness was first proposed by Galton (1879) and later demonstrated with the use of software applications that allow the creation of image averages. Averaging increasingly larger sets of male or female faces results in stimuli that are rated as more and more attractive (Langlois & Roggman, 1990). Sex‐typicality refers to feminine or masculine physical characteristics. Compared with male faces, female faces have, for example, higher cheekbones, fuller lips and larger eyes relative to total face size. Thus, most men perceive faces that emphasize these traits as more attractive (Cunningham, 1986; please note that for this and many other studies, the participants' sexual preferences had not been recorded). Interestingly, however, women, on average, fail to show a clear bias towards masculine male faces (Cunningham et al., 1990; Swaddle & Reierson, 2002). Finally, some studies have examined morphological differences between the left and right side of the face, showing that greater such differences are found to reduce facial attractiveness (Harun et al., 2023; Perrett et al., 1999).

Attractiveness beyond the face

While early research on attractiveness focused on the face, it became clear that other non‐verbal modalities are relevant also. Work beginning to look at these modalities revealed both similarities and differences. For example, research found that, like facial attractiveness, vocal attractiveness increased with voice averaging (Bruckert et al., 2010). Moreover, female voices were more attractive when their pitch was higher and hence more sex‐typical, whereas there was no clear preference for more masculine, deep voices (Apicella & Feinberg, 2009; Fraccaro et al., 2013). Yet, work on body odours added complexity. It showed that odour attractiveness is not biased towards body odour compounds, a proxy for averaging, relative to individual body odours (Fialová et al., 2018). Additionally, odour attractiveness ratings appear to rely in part on the extent to which a perceiver and an olfactory agent differ in the major histocompatibility complex (MHC), a set of genes responsible for coding surface proteins that present antigens to T cells, thus enabling adaptive immune system responses. In this context, women were found to rate body odours of MHC‐dissimilar men as more attractive compared with MHC‐similar men (Roberts et al., 2008; Wedekind et al., 1995; but see Havlicek et al., 2008), presumably because MHC variability confers greater fitness. Additionally, familiarity (Sorokowska et al., 2018), dietary influences (Havlicek & Lenochova, 2006), immune status (Schwambergová et al., 2023) and the female menstrual cycle (Gildersleeve et al., 2012) were found to affect body odour and to thus inform attractiveness. Whereas women, compared with men, reported a greater reliance on olfactory cues for mate choice, men, compared with women, reported a greater reliance on visual appearance (Havlicek et al., 2008).

The observation that different modalities contribute to attractiveness judgements has prompted questions as to whether the modalities encode information redundantly or whether they carry unique information about an individual's value as a mate or social partner. Work using faces and voices implied that there is some non‐verbal redundancy. For example, participants presented with a pair of faces matched a voice stimulus more readily with the more attractive face if the voice was also rated as attractive (Hughes & Miller, 2016). Furthermore, photographs of women with naturally high‐pitched voices were rated as more attractive than photographs of women with low‐pitched voices (Collins & Missing, 2003). Additionally, attractiveness ratings of photo and video material were positively correlated with attractiveness ratings of voices from the same individuals (Lander, 2008). Notably, however, existing work typically focused on opposite‐sex judgements and/or reported inconsistent results for male and female raters (Collins & Missing, 2003; Feinberg et al., 2005; Valentova et al., 2017; Wells et al., 2013).

To the best of our knowledge, there are only two studies that have examined non‐verbal redundancy involving body odours. One study looking at faces and odours found an association that depended on the raters' sex and, for female raters, on their menstrual cycle in that the association was present only during the most fertile menstrual phase (Rikowski & Grammer, 1999). Also here only opposite‐sex ratings were collected. The second and comparatively small‐scale study used pictures, videos and odours from male agents that were rated by a dozen female perceivers. The authors explored whether odours predicted visual attraction but only saw such an effect when correlating odour ratings with the residuals of a regression of picture against video ratings (Roberts et al., 2011). They argued that those residuals retained kinetic information and that the attractiveness of dynamic, movement‐related information co‐varies with that of odours.

Attractiveness beyond common preferences

Discussions regarding attractiveness have centered on common preferences that make a given non‐verbal agent appealing to a majority of opposite‐sex perceivers. The underlying idea has been that such preferences are functionally relevant in that they point to a high‐quality mate. In line with this, attractiveness has been linked to health, immunocompetence and reproductive value (Batres & Shiramizu, 2023; Thornhill & Gangestad, 1993, 2006; White et al., 2013) suggesting that being drawn to averageness, sex‐typicality and symmetry enhances one's chances of reproductive success. Notably, however, the support for such links is mixed and includes null results (Rantala et al., 2013; Schwambergová et al., 2023; Scott et al., 2013) as well as evidence for fairly complex associations between looks and mate quality (Mengelkoch et al., 2022). Thus, common preferences in attractiveness may not be as useful as initially thought.

In line with this, research implies that, in addition to common preferences, there may be perceiver‐specific biases that could inform attractiveness judgments. Indeed, some impression formation studies found perceivers look for both convergence with and divergence from their own personal characteristics, which together shape personal preferences. Evidence relevant for convergence comes from a phenomenon referred to as assortative mating (Hunt et al., 2015; Lutz, 1905), according to which we are biased towards others who are similar rather than dissimilar from us (e.g., in personality). Evidence relevant for divergence includes that certain bodily functions benefit from genetic variation leading us to prefer genetically different mates (Wedekind et al., 1995). Thus, who is attractive to whom depends not only on averageness, sex‐typicality or symmetry, but on personal preferences for a potential partner. To date, the relative importance of common vs. personal preferences in attractiveness has been rarely explored (for exceptions see Baxter et al., 2022; Eastwick & Hunt, 2014; Hehman et al., 2017).

Attractiveness beyond mate selection

Although attractiveness undoubtedly matters for mate selection, it is also relevant for platonic interactions with opposite‐ and same‐sex individuals (Little et al., 2011; White et al., 2013). This fact is best illustrated by the so‐called halo effect where physically beautiful individuals are associated with a range of positive personality traits that they may or may not possess (Albright et al., 1988; Dion et al., 1972). First established in Western cultures, this finding has since been replicated in non‐Western cultures (Batres & Shiramizu, 2023; Langlois et al., 2000). Studies conducted in Africa, Asia, Australia, Europe, the Americas and the Middle East have demonstrated correlations between attractiveness and how caring, confident, dominant, emotionally stable, happy, intelligent, responsible and trustworthy individuals were rated. Importantly, these correlations were the same for both male and female faces from raters of various sexes indicating that they affect us irrespective of romantic or reproductive interests (Batres & Shiramizu, 2023).

Adding to this, there is growing evidence that apart from sexual relationships, non‐sexual relationships matter for an individual's well‐being and reproductive success. Observations in hunter‐gatherer societies have highlighted the importance of same‐sex or platonic alliances for stress regulation and health (Holt‐Lunstad et al., 2010; Silk, 2007; Taylor, 2006) and for the number of living offspring (Page et al., 2017). Evidence from modern societies corroborates this. A survey asking participants to report their emotionally closest interaction partner found that 86.4% chose a same‐sex friend (Singleton & Vacca, 2007). Additionally, both the quantity and quality of platonic social relationships have been shown to positively predict adult well‐being (Holt‐Lunstad et al., 2010; Walton & Cohen, 2011) and healthy aging (Yang et al., 2016). Because they are shaped, to some degree, by mutual attraction (Cussigh et al., 2020; Hagerman et al., 2017; Little et al., 2011), the processes underpinning impressions of attractiveness become relevant for how we feel.

The present study

This study built on existing attractiveness research. Its participants were young adults who generated non‐verbal expressions in different modalities that would serve as evaluative stimuli. We took their facial photos, recorded upper body videos from which we retrieved separate audio and visual channels and collected axillary sweat samples. We invited a separate group of young adults to serve as non‐verbal perceivers. Perceivers were presented with the different stimuli and rated stimulus attractiveness. Our statistical analyses addressed three questions.

First, we pursued modality effects by exploring potentially unique and redundant modes of attractiveness signalling. We probed uniqueness by comparing the different non‐verbal conditions considering rating score variation and central tendencies. We probed redundancy by examining whether attractiveness in one condition positively predicted attractiveness in another condition. Our second question concerned the relative importance of common vs. personal attractiveness preferences. We used the statistical variance linked to agents to infer common preferences and the statistical variance linked to perceivers to infer personal preferences. Lastly, we asked about a possible function of attractiveness beyond mate choice. To pursue this function, we compared the attractiveness ratings for opposite‐ and same‐sex agents seeking to establish potential similarities.

METHODS

Participants

In line with a recent framework on human non‐verbal communication (Schirmer et al., 2025), a communicative process involves an active or emitting individual, called the agent, who produces (intentionally or non‐intentionally) non‐verbal cues and signals. Additionally, a perceiver observes the agent and in doing so perceives his or her cues and signals, which may influence the perceiver's own behaviour. Thus, in the present context, the term agent refers to participants who produced non‐verbal stimuli, while the term perceiver refers to participants who rated the same.

Although this study involved both agents and perceivers, we focused our power analysis on the required number of perceivers, as we found no software tool that readily supported the required multi‐level modelling in the context of ordinal data. We used the superpower package in R to estimate the perceiver sample size needed for a medium‐sized three‐way interaction (e.g., modality × agent sex × perceiver sex) in an ANOVA, with one factor being a between‐subject factor and the others being within‐subject factors. We focused on a medium‐sized effect because a small effect may not be very meaningful for everyday social interactions and because a large effect was deemed unlikely, especially as we were looking for an interaction effect. Indeed, being interested in contrasting the different non‐verbal modalities and examining to what extent modality effects may vary by agent and perceiver sex meant that we needed to ensure adequate power for the highest‐level interaction. Our analysis indicated that two perceiver sex groups with 34 participants each would yield 80% power.

This study recruited 61 participants to serve as non‐verbal agents and 71 participants to serve as perceivers. We recruited a little above our target to ensure that sexual preferences were fairly balanced across female and male perceivers. All participants had a minimum age of 18 years, had substantial German language competencies (level C1; considered relevant to producing and understanding the German sentence as described below) and reported an absence of psychological disorders (to mitigate related socio‐emotional processing biases; Bourke et al., 2010; Cortese et al., 2021). Additionally, we excluded participants who were not in good general health, as in when they had an infection or cold. Agents comprised 34 women aged 20.9 years (SD 3.8) and 27 men aged 22.8 years (SD 4.6). Perceivers comprised 37 women aged 21.7 years (SD 2.38) and 34 men aged 22.56 years (SD 3.25). On a scale of 1 (heterosexual) to 5 (homosexual), 22 female perceivers rated themselves as 1, twelve as 3 and three as 4. Twenty‐seven male perceivers rated themselves as 1, one as 2, three as 3, two as 4 and one as 5. A chi‐squared test for independence was performed comparing the counts for each rating category between male and female participants. This showed that both groups did not differ significantly in the distribution of their sexual orientation (X[4] = 7.99, p = .092).

All participants gave informed consent prior to participating in this study.

Materials

The present research questions and analyses are part of a larger project. For transparency, we describe the methods of this larger project in full and highlight what was irrelevant for the present purpose. Note that this concerns all questionnaires collected from agents and perceivers as well as some ratings collected from perceivers only (an overview of these data points is presented in the Table S1).

Participants, who were recruited as non‐verbal agents, performed the following tasks aimed at developing a stimulus set for subsequent evaluation. First, these participants were instructed to wash their arm pits with a neutral soap and to apply one collection pad under each armpit. They were then given a white T‐shirt to wear. Next, they entered an audio visual recording room where their photo was taken against a white background. For this photo, participants had to adopt a neutral, relaxed facial expression. Photos comprised of the participants face, hair and neck. Subsequently, they were asked to speak a German sentence (English translation ‘Nice to meet you. We'll definitely see each other again sometime.’) with the intention of greeting a stranger and to say an extended “Aaah” twice in succession, which were recorded on video. Please note that the sentence and utterance simply served as a vocal vehicle. Because they were identical across agents, the sentence/utterance content was not expected to explain variance in perceiver ratings. Videos captured a participant's head and upper body. During the next 10 min, participants were asked to walk up and down three flights of stairs 10 times as fast as possible so as to promote sweating for the olfactory samples. Finally, participants completed three personality questionnaires including the NEO‐FFI, SANB5 and SAM, which were excluded from the present analysis. At the end of this procedure, which took ~30 min, the experimenter collected the arm pads and placed them into separate zip‐bags that were stored at −27°C. All handling of arm pads was done using medical gloves.

After agent data collection was completed, photos were sized to a dimension of 1800 pixels in width and 1200 pixels in height. Additionally, videos recorded using different devices were processed with Adobe Premiere to have a comparable colour space. They were edited to show only the relevant utterances (Greeting: 5.1 s (SD 0.5); Aaah: 6.1 s (SD 1.4)) and their sound amplitude was normalized using the Normalize Mix Track function in Adobe Premiere so that it was comparable across conditions. Muted videos and the videos' audio extraction (i.e. the participant's voice) were stored separately and served later as video and audio stimuli, respectively.

Procedure

Participants serving as perceivers were presented with the stimuli of eight agents—four males and four females. Due to the number of modalities and rating tasks, a larger number of agents would have compromised the participants' engagement and motivation throughout the study, which already took 1 h. Different perceivers were assigned pseudo‐randomly to different agents so that across perceivers all agents were shown with comparable frequency (mean ~9, SD 1.6).

Perceivers rated the stimuli from all eight agents separately for each modality in the following order. First, they rated all agents with respect to body odour. Next came the facial photographs, followed by the audio only from both videos (in counterbalanced order), followed by both videos in muted mode (in counterbalanced order) and lastly, followed by both videos in unmuted mode (in counterbalanced order). This order was held constant so as to reduce the chances of a halo (or reverse halo) effect; that is, ratings being influenced by former potentially associated ratings (e.g. photo effects shaping how odours are being rated or audio–video effects shaping how individual audio or video stimuli are rated). Within each modality, the stimuli from each of the eight agents (irrespective of agent sex) were selected randomly for evaluation.

For each stimulus, participants responded to the following seven questions in the same order: Which emotion is most likely to be conveyed by this stimulus (joy, sadness, fear, surprise, disgust, anger and none)? Which emotion do you feel most when you perceive this stimulus (joy, sadness, fear, surprise, disgust, anger and none)? How attractive do you find this stimulus? How sexy do you find this stimulus? How much do you like this stimulus? How familiar are you with this stimulus? How much would you like to meet this person? The last five questions were rated on a scale from 0 (not at all) to 6 (very much). Additionally, when evaluating the body odours, participants reported odour valence (Does the odour trigger a negative or positive emotion? (−4 very negative to 4 very positive; 0 = neutral)) and semantic associations (Please write down three words that describe the odour you have just smelled). Of these questions, only ‘How attractive do you find this stimulus?’ entered data analysis.

Finally, participants completed the same three personality questionnaires as mentioned above, none of which were analysed here. The entire procedure took about 60 min.

Data analysis

To pursue the present research questions, we focused our analyses on the attractiveness ratings. Moreover, to ensure a balanced statistical design, we were forced to select either the spoken sentence or the ‘Aaah’ expressions from the audio and video stimuli (using both would have meant that voice/video ratings had twice the number of data points when compared with odour/photo ratings). Because the sentences were prosodically more complex and also more typical in the context of first impression formation, we opted for those.

All data were analysed in R (R Core Team, 2015). As the relevant dependent variable was ordinal, we used cumulative link mixed effects modelling as implemented in the ordinal package (Christensen, 2018). Moreover, we modelled mixed effects so as to account for and estimate the variance associated with both agents and perceivers. Accordingly, the intercepts of both agents and perceivers were included in the random effects term, which also allowed us to consider their nesting. Below, we only describe the fixed effects, which varied between the different models. The Wald test was used to compare a simple model with a more complex model that differed in only one effect (e.g. main or interaction effect) such that we could determine whether this effect was statistically significant.

Our independent variables included (i) modality (audio, photo, video, audio–video and odour), (ii) agent sex (opposite/same) and (iii) perceiver sex (male/female). We ran two different sets of ordinal regression analyses. One set compared the different modalities and thus had modality as a categorical fixed effect. The other set of analyses addressed possible positive associations between the modalities and so had the attractiveness ratings of one modality as the dependent variable and those of another modality as the fixed effect. In both sets of analyses, agent sex and perceiver sex as well as their interactions were entered as additional fixed effects.

In the context of the second set of analyses, we adopted an approach used elsewhere (Roberts et al., 2011) to obtain an estimate of largely kinetic information by regressing the photo ratings against the video ratings and using the residuals of that regression as a kinetic index. These residuals represent variance unaccounted for by the commonalities between photos and videos. One may reason that a main source for this variance was the movement that was absent from photos and present in videos. Note, however, the residuals may also represent other sources of variance; for example, that the videos showed, apart from the head and neck, the upper body (although we largely standardized this with the T‐shirts).

Our second set of analyses addressed the associations between different non‐verbal modalities. Here, we excluded the audio–video stimuli (i.e. un‐muted videos) because we wished to focus on the association between individual communication channels (i.e. separate audio and video recordings). In all, we tested relevant main and interaction effects of eight ordinal regression models—one for each combination of non‐verbal modalities (except between the video/photo conditions and the kinetic index as they were used to obtain this index). Note that for each of these regression models, we had to designate one modality as the dependent and the other as the independent variable and thus had two options of how to conduct a given analysis. We had no reasons to prefer one option over the other and simply picked one with the caveat that results might have been slightly different for the other option.

Note that this form of ordinal testing does not yield an effect size for main and interaction effects. Thus, we derived the AIC improvement associated with the added effect in model comparisons and report this improvement as an effect size (Burnham & Anderson, 2003). Values between 2 and 4 may be considered small effects, those between 4 and 10 medium effects and those above 10 large effects.

RESULTS

Modality differences in attractiveness as a function of agent and perceiver sex

As shown in Figure 1, we found overlap and divergence in how the modalities inform attractiveness ratings. Although the different non‐verbal conditions showed similar variation in attractiveness ratings, some were rated as more attractive than others, with audio–video stimuli being most attractive. Female and male perceivers differed in how they evaluated same‐sex agents but compared in how they evaluated opposite‐sex agents.

FIGURE 1.

Attractiveness ratings for the different non‐verbal modalities and as a function of the statistical factors agent sex (opposite/same) and perceiver sex (female/male). Dots represent individual agent/perceiver ratings. They are overlaid by box‐and‐whisker plots that are complemented by histograms. Dots are jittered for easier visibility. The thick horizontal grey line marks the scale midpoint (neither attractive nor unattractive).

In detail, we first employed the Fligner‐Killeen test with attractiveness as the dependent variable and modality as the independent variable. The test results implied that the variances for the different modalities were homogeneous (p = .226). Thus, perceivers used the modalities similarly when rating agent attractiveness. Next, attractiveness ratings entered an ordinal regression with modality (audio, photo, video, audio–video and odour), agent sex (same, opposite) and perceiver sex (female, male) as factors. This returned a significant effect of modality (X[4] = 150.97, p < .0001, ΔAIC = 142.97). The interaction between modality and agent sex (X[4] = 9.08, p = .059, ΔAIC = 1.08) and that between modality and perceiver sex (X[4] = 8.24, p = .083, ΔAIC = 0.24) failed to reach significance. However, the three‐way interaction reached significance (X[5] = 26.58, p < .0001, ΔAIC = 16.58).

We pursued these results by examining the modality by perceiver sex interaction for each level of agent sex. For same‐sex agents, the interaction effect was significant (X[4] = 30.27, p < .0001, ΔAIC = 22.27) and FDR corrected follow‐up tests revealed that women rated odour stimuli as least attractive and different from all other modalities (βs > 0.94, SEs < 0.22, ps < .0001). Next came audio, photo and video stimuli, which did not differ (ps > .145) and which were less attractive than audio–video stimuli (βs > 0.462, SEs < 0.209, ps < .033). Men rated photo stimuli as least attractive and different from all other stimuli (βs > 0.628, SEs < 0.242, ps < .023). Odour, audio, video and audio–video stimuli were all rated equally attractive (ps > .178). For opposite‐sex agents, the modality by perceiver sex interaction was non‐significant (p = .561). Hence, we explored the modality main effect and found that the odour condition was least attractive, followed by the photo, the audio, the video and the audio–video condition. All differences were significant (βs > 0.318, SEs < 0.161, ps < .039) with the exception of that between the audio and video condition (p = .374). For an overview of these results, please refer to Table 1.

TABLE 1.

Analysis of modality differences—overview of results.

Statistical effect	Attractiveness differences (lowest to highest)
Modality × Agent sex × Perceiver sex***
Same‐sex Agents
Modality × Perciever sex***
Women: Modality***	Odour < photo, audio, video < audio–video
Men: Modality***	Photo < odour, audio, video, audio–video
Opposite‐sex Agents
Modality × Perceiver sexn.s.
Modality***	Odour < photo < audio, video < audio–video

Note: ***p < .001; n.s., non‐significant effect (p > .1).

Modality redundancy as a function of agent and perceiver sex

Next, we probed the possibility that the manner in which individuals communicate across the different non‐verbal modalities reflects core person aspects that are at least partially redundantly coded. Notably, we found evidence for this from all modality combinations except for the combination of olfactory and vocal ratings and the combination of olfactory and photo ratings. Associations were fairly similar across different agent and perceiver sex combinations, albeit a slight bias towards larger/more consistent associations for opposite‐sex agents was observed (Figure 2).

FIGURE 2.

Associations between attractiveness ratings across the different non‐verbal modalities. The plotted values were derived from ordinal mixed modelling whereby the attractiveness ratings of one modality were regressed against the attractiveness ratings of another modality. For illustration only, separate models were computed for each combination of agent and perceiver sex, showing the beta coefficient (^† p < .1, *p < .05, **p < .01, ***p < .001). A more detailed figure with individual data points and regression lines is provided in the Figure S1.

In detail, an analysis with olfactory ratings as the dependent variable and auditory ratings as the independent variable yielded only non‐significant results (ps > .282).

An analysis with olfactory ratings and photo ratings revealed a significant interaction between photo ratings and agent sex (X[1] = 4.86, p = .027, ΔAIC = 2.86; all other ps > .196). Follow‐up regressions yielded a weakly positive but non‐significant association between olfactory and photo attractiveness for opposite‐sex agents (β = 0.171, SE = 0.092, p = .064) but not for same‐sex agents (p = .555).

Next, we examined the association between olfactory and video ratings. Here, we observed a significant video main effect (X[1] = 4.27, p = .039, ΔAIC = 2.27) while the interaction between video ratings and agent sex failed to reach significance (X[1] = 2.82, p = .093, ΔAIC = 0.82). All other effects were non‐significant (p > .608). The video main effect was driven by a weakly positive association between olfactory and video attractiveness (β = 0.146, SE = 0.07, p = .038).

Looking at the association between olfactory ratings and the kinetic index revealed a significant kinetic main effect (X[1] = 7.69, p = .005, ΔAIC = 5.7), whereas all other effects were non‐significant (ps > .681). The kinetic effect implied a positive association between olfactory and body motion attractiveness (β = 0.239, SE = 0.086, p = .005).

An analysis with audio ratings as the dependent variable and photo ratings as an independent variable returned a significant photo rating effect (X[1] = 44.26, p < .0001, ΔAIC = 42.26). The interaction between photo ratings and perceiver sex (X[1] = 3.35, p = .067, ΔAIC = 1.35) and all other effects were non‐significant (ps > .756). The photo rating effect was due to a positive association between auditory and photo attractiveness (β = 0.517, SE = 0.078, p < .0001).

Similar results emerged when regressing video against audio ratings. Again, the video main effect was significant (X[1] = 50.96, p < .0001, ΔAIC = 48.96) with all other effects being non‐significant (ps > .11). Like photo attractiveness, video attractiveness predicted auditory attractiveness positively (β = 0.552, SE = 0.076, p < .0001).

An examination of the association between auditory ratings and the kinetic index was also significant. There was a kinetic main effect (X[1] = 9.94, p = .001, ΔAIC = 7.94) and a significant interaction between kinetic index, agent sex and perceiver sex (X[1] = 5.11, p = .024, ΔAIC = 3.11). For same‐sex agents, the interaction between the kinetic index and perceiver sex was significant (X[1] = 4.77, p = .029, ΔAIC = 2.77) indicating that the kinetic index predicted auditory attractiveness positively in male (β = 0.677, SE = 0.24, p _FDR = .01) but not female perceivers (p _FDR = .963). For opposite‐sex agents, the two‐way interaction was non‐significant (p = .389) indicating that the kinetic index predicted auditory attractiveness positively independently of perceiver sex (β = 0.285, SE = 0.137, p = .038, ΔAIC = 2.1).

Finally, we regressed the photo ratings against the video ratings, finding a strong photo main effect (X[1] = 219.94, p < .0001, ΔAIC = 219.94) with all other effects being non‐significant (ps > .15). Not surprisingly, photo attractiveness predicted video attractiveness strongly positively (β = 1.23, SE = 0.088, p < .0001; Figure 2).

An overview of these results is shown in Table 2.

TABLE 2.

Analysis of modality redundancy—overview of results.

Model	Statistical effect
Odour ~ Audio × Agent sex × Perceiver sex	n.s.
Odour ~ Photo × Agent sex × Perceiver sex	Photo × Agent sex*
	Same‐sex Agents: Photo n.s.
	Opposite‐sex Agents: Photo†
Odour ~ Video × Agent sex × Perceiver sex	Video*
	Video × Agent sex†
Odour ~ Kinetic × Agent sex × Perceiver sex	Kinetic**
Audio ~ Photo × Agent sex × Perceiver sex	Photo***
	Photo × Perceiver sex†
Audio ~ Video × Agent sex × Perceiver sex	Video***
Audio ~ Kinetic × Agent sex × Perceiver Sex	Kinetic**
	Kinetic × Agent sex × Perceiver sex*
	Same‐sex Agents: Kinetic × Perceiver sex*
	Women: n.s.
	Men: Kinetic *
	Opposite‐sex Agents: Kinetic *
Video ~ Photo × Agent sex × Perceiver sex	Photo***

Note: ^† p < .1, *p < .05, **p < .01, ***p < .001; n.s., non‐significant effect (p > .1).

Variance accounted for by agents and perceivers as a function of agent and perceiver sex

In the last step, we explored the variance in the rating data that was accounted for by agents and perceivers so as to gauge the importance of common (e.g. averageness) and personal preferences. Moreover, we sought to probe whether one source of variance was more important than the other. Towards this end, we bootstrapped the rating data (without replacement, 24 female and 24 male perceivers, 1000 samples) and subjected the bootstrapped samples to ordinal regressions with the rating scores as the dependent variable and the intercepts of agents, perceivers and modality as the random effects (no fixed effects were modelled). For each sample, we conducted four separate regressions, one for each combination of agent and perceiver sex. From each regression, we derived the ICCs of agent and perceiver effects by dividing their respective variance estimates by the total estimated variance. We then took the difference between the agent and the perceiver ICCs and obtained 95% confidence intervals for that difference. As illustrated in Figure 3, for same‐sex agents, the perceiver ICC was significantly greater than the agent ICC in both female and male participants (the CI excluded 0). For opposite‐sex agents, however, the two sources of variance differed in neither female nor male participants (the CIs included 0).

FIGURE 3.

ICC results from the bootstrapped regressions described in the text. Illustrated on the left are the mean ICCs and their 95% confidence intervals for the different combinations of agent and perceiver sex. The ICCs associated with the agent random effect are shown in green, whereas the ICCs associated with the perceiver random effect are shown in orange. Illustrated on the right are the ICC differences between agents and perceiver random effects. Differences were obtained by taking the ICCs associated with the perceiver random effect and subtracting the ICCs associated with the agent random effect. From this we derived both means and 95% confidence intervals showing that the differences were significant for same (excluded 0) but not opposite sex (included 0) ratings. Expressed differently, for same but not opposite‐sex ratings were personal preferences more relevant than common preferences.

Note that we also attempted to examine the ICCs for each modality separately. Here, however, a bootstrapping approach was not feasible as the resulting data sets were too small for modelling and led to convergence failures. Therefore, we could only explore the modalities for the full set of agents and perceivers and were unable to derive statistical estimates concerning condition differences. For the interested reader, we report the modality‐specific ICCs in the Figure S2.

DISCUSSION

Much research has tackled the factors determining our attraction to other people. However, this research has been largely focused on faces, common preferences and opposite‐sex processes in the context of mate selection. Here, we sought to extend this work by combining the traditional facial photos with dynamic auditory, visual as well as olfactory person information. Moreover, we considered the relative importance of common vs. personal preferences for an agent's perceived attractiveness. Finally, both agent and perceiver sex varied, allowing us to determine whether the processes that shape attraction are specific to opposite‐sex encounters or generalize to same‐sex encounters, which, for most individuals, are largely platonic in nature.

Attractiveness is multimodal and partially redundant

We approached the issue of multimodality in two ways. Specifically, we asked how the modalities differ in impression formation and whether they provide redundant information. With respect to differences, we found them to be significant but small. Although participants used the entire 7‐point rating scale ranging from not at all to very much attractive, mean ratings differed by less than a point (i.e. 0.8) between the different non‐verbal conditions. Average modality attractiveness ranged from somewhat below the scale midpoint to the scale midpoint. Least attractive were odour stimuli followed by photos. Audio and visual stimuli ranked in the middle, while combined audio–video stimuli were clearly most attractive. These results are largely compatible with a former observation of voices being more attractive than static and moving faces (Lander, 2008). Additionally, the fact that the dynamic, multimodal condition was most attractive corroborates the importance of rich, time‐varying information for attractiveness and, perhaps more generally, for impression formation. Indeed, multimodal integration may play a special role in attractiveness whereby the positive impressions we gain from individual non‐verbal channels add up or amplify each other. Perhaps perceiving an attractive voice helps identify attractive facial features and vice versa, thus producing a sort of halo effect. Further research will have to probe this possibility.

With respect to modality redundancy, we found that attractiveness positively correlated between most of the individual modalities we explored. This corroborates past research pursuing the attractiveness of opposite‐sex individuals with photos and voices (Collins & Missing, 2003; Lander, 2008; Valentova et al., 2017; Wells et al., 2013) as well as with videos on the one hand and facial photos, upper body photos and voice recordings on the other hand (Saxton et al., 2009). Our findings also concur with an original attempt to link olfactory with body motion attractiveness in female agents by using a regression‐based kinetic index derived from the ratings of male perceivers (Roberts et al., 2011). Importantly, the present results extend previous research by showing that these effects apply to same‐sex attractiveness and are thus important beyond mate selection. Moreover, they reveal additional statistical relationships that have, to the best of our knowledge, not been reported elsewhere. This includes positive associations between the kinetic index and voices as well as between olfactory stimuli and muted videos.

Association strength varied between the different modalities. They were strongest between the visual and audio stimuli and smallest between these modalities and the body odours, where, in some cases, they were non‐significant (i.e. photo, audio). Body odours were most consistently linked to the kinetic index thought to reflect body motion. That the different non‐verbal modalities relate to each other differently could have different reasons. One reason might be that attractiveness ratings are easier in some than in other modalities such that ratings are less noisy, leading to more robust associations. Indeed, one may argue that we are most likely to explicitly judge another's looks and the sound of their voice such that these judgements are most likely to concur. A second reason might be that the different modalities convey different types of information such that those conveying similar types more readily correlate in attractiveness. In support of the latter possibility is research showing that a person's face, body built and voice change only slowly with age, while body odours are more dynamic as they depend on ongoing metabolic processes and are determined, apart from inborn properties (e.g. MHC), by nutrition (Havlicek & Lenochova, 2006), hormonal status (Gildersleeve et al., 2012; Schäfer et al., 2020) and acute inflammation (Olsson et al., 2014). Similarly, movement is a dynamic modality that is impacted by nutrition (Fritzen et al., 2019; Johnson & Leck, 2010), hormones (Casey et al., 2014; Julian et al., 2017; Lee et al., 2017) and inflammation (Lasselin et al., 2020). Thus, the redundancy between vision and audition may be due to stable, shared person characteristics and support person identity recognition, which humans find comparatively more difficult based on body odour (Lundström & Jones‐Gotman, 2009) or movement (Loula et al., 2005). By contrast, the redundancy between body odour and movement may be due to corresponding dynamic changes in these modalities. Being less relevant for familiarity, these modalities may more readily support insights into aspects of a person's state that are relevant for attractiveness.

Having established that attractiveness is redundant across the different non‐verbal channels, one may ask why such redundancy exists. What might be the functional significance of conveying the same information across different non‐verbal channels? While we can only speculate, one possibility is that redundancy facilitates affiliative processes and bonding. If the different non‐verbal modalities were to convey different information such that an individual is attractive in one but not the other modality, finding partners that maximize attraction would be challenging. By contrast, redundancy enables us to identify attractive individuals through a convergence across the senses and thus biases social interactions that promise to be rewarding and beneficial.

Attractiveness depends on common and personal preferences

Past research has identified a few common preferences underpinning attractiveness: averageness, sex‐typicality and symmetry (Gangestad & Scheyd, 2005; Little et al., 2011; Thornhill & Gangestad, 1993; Walster et al., 1966). The central question has been which non‐verbal agents are attractive to the majority of perceivers, and this has been tackled by ranking non‐verbal agents based on the mean ratings of perceivers. Deviating from this approach, a few more recent studies have considered attractiveness as an individual process that, apart from the identified common preferences, is subject to personal preferences (Baxter et al., 2022; Eastwick & Hunt, 2014; Hehman et al., 2017). Here, the variance of attractiveness ratings obtained after in‐person interactions was examined with respect to the contribution of agents and perceivers. The results revealed a bias towards perceivers, implying a greater role of personal preferences compared with common preferences (Eastwick & Hunt, 2014; Hehman et al., 2017) especially with a longer acquaintance between agents and perceivers (Eastwick & Hunt, 2014). Indeed, personal preferences linked to compatibility with a potential partner have been found relevant for predicting later romantic desire and dating (Baxter et al., 2022).

Our study builds on this work while taking a more analytic approach focused on isolated non‐verbal impressions before an actual interaction, thus excluding the influence of an agent's response to a perceiver. Moreover, we broadened the existing pursuit of mate selection to include same‐sex social possibilities. Deviating from previous results, we found that the variance in attractiveness ratings explained by perceivers did not differ from that explained by agents in opposite‐sex ratings. Yet, a significant bias towards perceivers was shown for the ratings of same‐sex agents. Based on these results, we reason that a personal bias of opposite‐sex impressions may be absent at the beginning of an encounter when another is first perceived and emerge only subsequently with the reciprocity of social exchange.

Together, the present observations help qualify earlier reports and highlight the importance of separately considering the various stages of an interaction or relationship and of exploring both same‐sex and opposite‐sex impressions. Moreover, they raise the question as to why personal preferences matter so much for whom we find attractive? Building on the similarity attraction hypothesis (Izard, 1960) as well as research highlighting a relevance of genetic variation (Wedekind et al., 1995), we reason that this is because the perceivers themselves vary substantially in their personal characteristics and that these characteristics matter for whom they find attractive. Who they are, their biological make‐up and life experiences make some individuals more suitable as lovers or friends than others (Montoya et al., 2008; Wilson et al., 2016) and the processes that give rise to attraction help identify these individuals.

Attractiveness matters for all social encounters

Attractiveness research has, for a long time, been motivated by questions concerning mate choice and pair bonding and thus focused on the role of non‐verbal effects in the context of opposite‐sex encounters (Caton et al., 2024; DuVal et al., 2023; Gangestad & Scheyd, 2005; Scheller et al., 2024). In line with the presumed importance of such encounters, the present study identified a few sex‐specific effects. With respect to modality differences, female and male perceivers compared in their ratings of opposite‐sex stimuli but differed in their ratings of same‐sex stimuli. Here, female perceivers rated odours and male perceivers rated photos as least attractive. With respect to modality redundancy, there was a weak tendency for odour and photo attractiveness to correlate for opposite but not same‐sex agents. Additionally, an association between vocal and kinetic attractiveness showed independently of sex for opposite‐sex agents but in only male perceivers for same‐sex agents. Lastly, there were significant differences in how perceiver and agent variance accounted for the rating data. Whereas both were equally important in the judgement of opposite‐sex individuals, perceiver variance was significantly more important than agent variance in the judgement of same‐sex individuals.

These differences between same‐ and opposite‐sex ratings suggest that attractiveness processes differ in strength and/or quality as a function of partner sex; thus, potentially serving somewhat specialized roles. Indeed, the modalities may differ in their relevance for identifying potential friends when compared with optimal reproductive partners. Olfactory cues, for example, may be more important for the latter, especially for male perceivers. What seems most striking, however, is the personal preference bias we observed for same‐sex ratings that was absent for opposite‐sex ratings. Perhaps personal preferences provide more clues as to agent–perceiver compatibility when dealing with non‐reproductive life tasks. By contrast, common and personal preferences may be equally but differently relevant in the context of creating and raising offspring. Common preferences may relate more to an individual's genetic endowment, whereas personal preferences may relate more to their capacity for joint parental care. Clearly, both need to be balanced for a successful outcome.

Notably, most of the present differences between opposite‐ and same‐sex attraction were small and complemented by many similarities. Indeed, the majority of our effects showed irrespective of whether agents might solicit platonic or sexual/romantic interests. These results align with other research delineating a role of attractiveness beyond mate choice (Batres & Shiramizu, 2023; Langlois et al., 2000) and support the idea that attraction not only guides the selection of sexual partners but also informs our pursuit of social exchanges more broadly.

Future directions

The present study has yielded a number of novel findings, prompting several questions for future research. First, with respect to modality, the kinetic index results point to a role for motion in how we find others attractive. Moreover, they suggest that how someone moves reveals information about their suitability as a sexual and platonic partner. Further research must test this possibility by involving other means to capture and present motion, such as point‐light displays (Atkinson et al., 2004). Moreover, additional work is needed to take a closer look at how the different modalities diverge and converge and the processes by which attractiveness increases and, perhaps, decreases as the modalities combine (e.g. audio/video vs. photo/odour). Second, with respect to the role of common vs. perceiver preferences, the current data show the latter to be as if not more important than the former. This finding raises questions concerning the function of perceiver preferences. We speculated here that they may serve to find suitable social partners, which should be explored in follow‐up research examining whether perceiver effects engender mutual attraction (i.e. perceivers prefer agents who are likely to prefer them in turn) and how attraction predicts interaction outcomes. Finally, this study reports a clear role for attractiveness in both opposite and same‐sex encounters while identifying potential qualitative and quantitative differences. More research is needed to understand these differences and to reveal their functional basis.

CONCLUSIONS

In sum, the present data yield unprecedented insights into how the modalities as well as common and personal preferences contribute to perceptions of attractiveness in both opposite‐ and same‐sex encounters. We found much variation in the attractiveness judgements of vocal, facial, bodily and olfactory stimuli with judgement means being lowest for body odours. Notably, multimodal audio‐visual stimuli were rated as more attractive than single modality stimuli pointing to an importance of multimodal integration. Statistically pairing the different modalities yielded a number of attractiveness correlations suggesting that core person aspects are coded redundantly thus creating clear biases in interpersonal preferences. While some aspects of these biases are shared between individuals, others vary in line with the notion that ‘beauty is in the eye of the beholder’. Together, an agent's non‐verbal qualities and their perceived attractiveness promote both opposite‐ and same‐sex interactions that promise positive interaction outcomes.

AUTHOR CONTRIBUTIONS

Annett Schirmer: Conceptualization; writing – original draft; visualization; formal analysis; writing – review and editing. Marcel Franz: Data curation; writing – review and editing; investigation; methodology; project administration; software; resources. Lea Krismann: Data curation; writing – review and editing; investigation. Vanessa Nöring: Data curation; writing – review and editing; investigation. Marlen Große: Data curation; writing – review and editing; investigation. Mehmet Mahmut: Conceptualization; writing – review and editing; data curation; project administration; methodology. Ilona Croy: Conceptualization; writing – review and editing; supervision; project administration; funding acquisition; methodology; resources.

Supporting information

Figures S1‐S2.

Table S1.

Schirmer, A. , Franz, M. , Krismann, L. , Nöring, V. , Große, M. , Mahmut, M. , & Croy, I. (2025). Attraction in every sense: How looks, voice, movement and scent draw us to future lovers and friends. British Journal of Psychology, 116, 684–701. 10.1111/bjop.12787

DATA AVAILABILITY STATEMENT

The data of this study are available at https://osf.io/typ4x/?view_only=73d7079444d847248eb5a6963ea04870.