Human recognition of feline stress-related behavioral states from visual cues depends on observer characteristics

Abstract

This study investigates humans’ ability to recognize cats’ stress-related behavioral states expressed through visual body language, including facial expressions, posture, and tail position. A total of 1,950 participants evaluated 12 videos of cats displaying three behavioral states—relaxed, tense, and fearful—and reported their perceived state. We examined whether recognition accuracy was influenced by individual observer characteristics, including age, gender, and prior cat ownership. Response accuracy exceeded chance level (33%) but remained relatively low, indicating that the task was challenging. No significant main effect of behavioral state was observed. In contrast, individual observer characteristics significantly shaped performance: participants identifying as female and those with prior cat ownership showed higher accuracy. Age also had a small but reliable negative effect, with accuracy gradually decreasing across adulthood. Overall, these findings indicate that humans show a general difficulty in detecting feline stress based on visual cues alone. Instead, recognition performance appears to be driven primarily by observer-related factors. This highlights the complexity of human–cat communication and underscores the importance of individual features in interpreting feline behavior. Improving humans’ ability to detect subtle visual indicators of feline stress may help foster more positive interactions and support companion animal welfare.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-44812-x.

Introduction

The domestic cats have been living with humans for around 10,000 years¹. They are extremely popular companion animals in households around the world. In the European Union, it is estimated that there are 113 million cats, more than the estimated 92 million dogs². Despite this popularity, research on cat behavior and welfare is still limited, and even more the research on the mechanisms that regulate cat-human interactions². Cats are facultative social animals³: they can establish social relationships with members of their own species (conspecifics) as well as with individuals of other species (heterospecific) such as humans⁴. The ability to perceive the emotions of other individuals plays a central role for animals living in a social context⁴, supporting communication among individuals. Emotions regulate social interactions, including those between humans and cats. A substantial body of research has demonstrated that cats are capable of perceiving and interpreting human emotional cues conveyed through auditory, visual, and olfactory channels^4–9. For instance, cats have been shown to spend more time in contact with their owners when the latter express positive emotions, such as happiness, compared to negative emotions like anger¹⁰. Furthermore, cats use emotional information expressed through human facial expressions to guide their behavior, particularly in situations involving novel or potentially threatening stimuli. This behavior, known as social referencing, is characterized by cats looking at their owners’ faces and adjusting their responses accordingly. When owners exhibit negative emotional expressions, cats are more likely to remain vigilant, whereas positive emotional cues are associated with more relaxed and stationary behavior⁶. Similarly, Quaranta et al.⁴ demonstrated that cats are capable of functionally recognizing human emotions such as anger and happiness by matching facial expressions with corresponding non-verbal vocalizations. Cats exposed to angry expressions and vocalizations exhibited more stress-related behaviors than those exposed to happiness cues, suggesting an integrated perception of multimodal emotional signals⁴. More recently, d’Ingeo et al.⁵ provided evidence that cats also respond to human emotional cues conveyed through olfaction. When exposed to human body odors associated with different emotional contexts (fear, happiness, physical stress, and neutral), cats showed higher stress levels in response to the odor linked to fear⁵. Collectively, these findings support the notion that cats actively engage in complex emotional communication with humans.

To date, few studies have analyzed the ability of people to perceive and interpret cats’ emotions. Recent literature has shown that humans are capable of perceiving feline emotions through auditory signals, particularly meowing. For instance, Prato-Previde et al.¹¹ found that people can distinguish between meows produced in different everyday contexts, such as when a cat is waiting for food, being brushed, or experiencing isolation—with the food-related meow being the most easily recognized. Similarly, Nicastro et al.¹² reported that humans more accurately identified meows associated with positive emotional states, such as initiating interaction or requesting food, compared to those linked to negative contexts like conflict, frustration, or distress. Supporting these findings, de Mouzon et al.¹³ also observed that meows associated with positive emotional states —such as anticipation of food, seeking interaction, or general contentment— were more readily recognized by human listeners than those linked to negative emotional states (i.e. discontentment). Together, these results suggest that humans are more attuned to feline vocalizations conveying positive affect.

In addition to auditory cues, studies have shown that people can recognize emotions expressed by cats through their facial expressions, although the success rate is generally low¹⁴. One possible explanation for these low recognition rates lies in the evolutionary history of the domestic cat. As a species that evolved primarily as a solitary predator—and for which the domestication process did not select for enhanced communicative abilities—cats lack the complex visual signaling repertoire typical of social carnivores³. Social predators often display rapid and conspicuous changes in emotional state that can be readily identified by group members to coordinate shared activities¹⁵. In contrast, feline visual communication consists of less complex and subtle cues, many of which are associated with defensive or agonistic contexts³. These signals are distributed across the entire body, integrating posture, facial expressions and tail position³. Such multi-component visual patterns suggest that interpreting cats’ emotions may require attending to whole-body configurations rather than to the sole facial expression.

Overall, the human ability to recognize feline emotions appears to be influenced by individual factors, including prior experience with cats and the observer’s age, gender and level of empathy toward cats^11,14. Dawson et al.¹⁴ reported that humans’ ability to recognize cats’ emotional facial expressions tends to decline with increasing age. Furthermore, women and individuals with professional veterinary experience (e.g., technicians and veterinarians) demonstrate greater accuracy in recognizing feline emotions compared to those with only personal or no experience. Interestingly, personal experience alone, such as simply owning a cat, showed limited effectiveness in improving recognition ability¹⁴. Supporting these findings, Prato-Previde et al.¹¹ showed that both personal experience and gender affect the recognition of feline emotional vocalizations: individuals with greater experience and women were more accurate at identifying meows produced in negative contexts such as brushing and isolation. However, they reported that cat owners outperform non-owners in distinguishing between different vocalizations, suggesting that daily interaction enhances the ability to interpret the emotional content of feline vocal signals. Furthermore, higher levels of empathy toward cats were associated with better recognition of meows emitted during isolation¹¹.

Considering this evidence, the present study aimed to further investigate humans’ ability to recognize cats’ stress-related behavioral state by examining visual signals expressed through the entire body. Extending previous work such as Dawson et al.¹⁴, which focused primarily on facial expressions, our approach incorporates full-body language—including body posture, facial expressions, and tail position—providing, for the first time, a more comprehensive and ecologically valid framework to study human perception of feline emotional communication. This study examines specific behavioral states (relaxed, tense, fearful) and assesses the potential influence of human gender, age, and ownership on recognition accuracy. These behavioral states were selected to represent distinct points along a continuum of stress intensity, ranging from relaxed to increasing levels of discomfort (tense) and fear, as described in established ethological frameworks of feline behaviour³. This choice allowed us to investigate humans’ sensitivity to progressively more intense signs of emotional discomfort that are highly relevant for cat welfare and everyday human–cat interactions.

Results

The final GLMM included behavioral states (relaxed, fear, tense), participant gender (female, male), previous experience with cats (ownership vs. no ownership), and age as fixed effects, with random intercepts for participants and stimulus videos.

The fixed intercept of the logistic mixed-effects model was significantly positive (β = 0.76, SE = 0.37, z = 2.09, p < 0.05), corresponding to a baseline level of performance above chance (33% correct responses on the probability scale). This indicates that, at the reference levels of the predictors, participants’ response accuracy was reliably above chance.

No significant effect of behavioral states was observed, as neither fear nor tense differed from the relaxed condition (p > 0.05), indicating that response accuracy did not vary statistically across behavioral expressions when controlling for age, gender, and experience with cats.

Gender significantly predicted response accuracy (β = 0.259, SE = 0.034, z = 7.69, p < 0.001), with female participants showing higher odds of providing a correct response than males (OR = 1.30, 95% CI [1.21, 1.38]; Fig. 1).

Fig. 1.

Mean proportion of correct responses by gender. Bars represent the observed mean accuracy expressed as percentage. Error bars represent ± 1 standard error of the mean. The dashed horizontal line indicates chance-level performance (33%). ***p < 0.001.

Previous experience with cats also had a significant effect (β = −0.424, SE = 0.035, z = − 12.15, p < 0.001), with cat ownership associated with higher odds of a correct response (OR = 1.53, 95% CI [1.43, 1.64]; Fig. 2).

Fig. 2.

Mean proportion of correct responses by cat ownership. Bars represent observed mean accuracy expressed as percentages. Error bars indicate ± 1 standard error of the mean. The dashed horizontal line denotes chance-level performance (33%). ***p < 0.001.

Age showed a negative association with performance (β = −0.0041, SE = 0.00112, z = − 3.65, p < 0.001), corresponding to a decrease in the odds of a correct response with increasing age (OR = 0.996, 95% CI [0.994, 0.998]; Fig. 3).

Fig. 3.

Predicted probability of a correct response (in %) as a function of centered age, estimated from the generalized linear mixed-effects model. Shaded areas represent 95% confidence intervals. The dashed horizontal line indicates chance-level performance (33%). ***p < 0.001.

Discussion

Overall response accuracy was 54%, exceeding chance-level performance (33%) but remaining relatively low, indicating that the task was challenging for participants. Descriptively, accuracy varied across behavioral expressions, with relaxed states identified more accurately (67%) than tense (49%) and fear (49%). However, performance was far from ceiling even for relaxed stimuli, suggesting that none of the behavioral categories was clearly or consistently distinguishable. Analysis of confusion patterns revealed systematic misclassifications: fear expressions were confused at comparable rates with both tense (29%) and relaxed states (22%), whereas tense expressions were more frequently misclassified as relaxed (39%); similarly, relaxed expressions were often confused with tense (25%). Together, these patterns point to a general difficulty in discriminating between behavioral states based on visible cues. In addition, response accuracy varied across individual stimulus videos, with variability observed for tense (S.D.= 22%), fear (S.D.= 19%) and relaxed (S.D.= 15%) expressions. This stimulus-level variability suggests that some videos were more difficult to interpret than others, potentially reflecting differences in behavioral expression intensity or individual characteristics of the cats depicted. Together, these observations motivated the use of a mixed-effects modelling approach to simultaneously account for participant-level and stimulus-level sources of variability. Importantly, when these descriptive patterns were formally tested using a generalized linear mixed-effects model that accounted for age, gender, ownership, and random variability associated with participants and stimulus videos, no significant main effect of behavioral states emerged. This indicates that the subtle descriptive differences observed across behavioral states did not translate into robust effects once other sources of variability were controlled for. The absence of a significant stress-related behavioral effect is consistent with previous studies reporting limited human accuracy in recognizing feline emotional-related states. Dawson et al.¹⁴ reported low performance when participants were asked to identify emotional valence from videos displaying cats’ facial expressions, while Prato-Previde et al.¹¹ observed similar difficulties when emotional information was conveyed through vocalizations embedded in contextual scenarios. Importantly, de Mouzon et al.¹³ showed that recognition accuracy increases only when emotional cues are presented in a bimodal format combining visual and vocal signals, whereas visual-only stimuli—particularly for negative emotional states—remain more difficult to decode. Taken together, these findings suggest that when humans rely on visual cues alone, emotion-specific discrimination is inherently challenging. Our result is particularly relevant in the context of everyday human–cat interactions and animal welfare. The behavioral states examined in the present study reflect different levels along a stress-related continuum, and, outside of highly intense situations, such states are typically not accompanied by vocalizations. In daily contexts, humans are therefore required to rely primarily on visual behavioral cues to detect early or moderate signs of feline stress. The observed difficulty in discriminating these states based on visual information alone highlights a critical limitation in human sensitivity to feline stress-related behaviors, with potential consequences for timely recognition of discomfort and for the prevention of negative human–cat interactions. These findings underscore the need for increased awareness and education regarding subtle visual indicators of feline stress to promote more effective communication and improved welfare in domestic cats. Within this framework, the absence of robust behavioral state–specific effects in the present study suggests that recognition performance is shaped more strongly by individual observer characteristics than by the behavioral category itself. This interpretation is supported by the significant participant-level effects observed in the mixed-effects model, indicating substantial inter-individual variability in sensitivity to feline behavioral cues.

Among individual observer characteristics, gender emerged as a significant predictor of performance, with female participants showing higher accuracy than male participants. This finding is consistent with a substantial body of literature reporting gender-related differences in emotion recognition, both in human–human¹⁶ and human–animal contexts^11,14,17, with females typically outperforming males in tasks involving the decoding of emotional and social cues. Similar gender effects have been reported in the recognition of facial emotional cues in cats^11,14 and dogs¹⁷, as well as in the perception of emotional signals in humans¹⁶. More recently, Hiisivuori et al.¹⁸ reported no gender differences for valence recognition in wild animals suggesting that gender-related advantages may not generalize uniformly across species or contexts. It is worth noting that gender in the present study was assessed through self-reported gender identity, while no information was collected regarding biological sex. Although a small number of participants identified as non-binary, their numerosity did not allow for a separate statistical examination of this group. Future studies including larger and more diverse gender identities, as well as measures of biological and hormonal factors, may help to further clarify the mechanisms underlying gender-related variability in interspecific emotion recognition.

In addition to gender, ownership significantly influenced accuracy in recognizing cat behavioral states. In line with previous evidence showing that experience with a given species can enhance emotion recognition abilities¹⁹, our results indicate that participants who had owned a cat were more accurate than those without such experience. This finding is consistent with studies by Ellis et al.²⁰ and Prato-Previde et al.¹¹, who reported higher accuracy in interpreting the communicative context of meows when the cat was familiar (i.e. a pet) rather than unfamiliar. Living with a cat may therefore facilitate increased sensitivity to subtle behavioral cues through repeated exposure and learning in everyday interactions. However, the relationship between pet ownership and emotion recognition is not necessarily straightforward. Several studies in both cats¹⁴ and dogs^17,21 suggest that ownership alone does not invariably translate into superior emotional state recognition skills. For example, dog ownership does not consistently improve the recognition of fear or anxiety²², nor does it reliably enhance the interpretation of behavioral or facial cues from static images^17,21. Importantly, a stronger and more consistent effect has been reported for professional experience with animals, such as that of veterinarians or animal behavior specialists, who typically outperform lay individuals in emotion recognition tasks. This pattern has been observed across species, including dogs²³ and macaques²⁴. Together, these findings suggest that structured experience and sustained, attentive exposure—rather than pet ownership per se—may be key factors in the development of interspecific “emotional”-related cues recognition skills. Future studies are therefore needed to disentangle the relative contributions of everyday ownership, targeted training, and professional expertise in shaping human sensitivity to cat visual emotional-related behavioral cues.

Age also contributed to individual variability in performance, with increasing age associated with a small but reliable reduction in accuracy when recognizing feline behavioral states. Although the magnitude of this effect was modest, it aligns with a broader literature documenting gradual age-related declines in emotion recognition across adulthood. Olderbak et al.²⁵ reported a clear age-related gradient in human emotion recognition performance: participants aged 15–30 years achieved the highest accuracy. Performance declined progressively with age, with individuals aged 46–60 years scoring lower than younger groups but higher than those aged 60 years and above, who showed the lowest accuracy overall. These findings are consistent with previous research showing that younger adults (18–24 years) generally outperform older adults in recognizing human emotions^26,27. Comparable age-related patterns have also been observed in emotion recognition in cats. Specifically, Dawson et al.¹⁴ found younger adults were more accurate than middle-aged adults (45–64 years) in interpreting cats’ facial expressions, suggesting that age-related changes in emotion recognition extend beyond human–human interactions. Declines in emotion recognition are especially pronounced in later adulthood. Individuals aged 65–83 years show reduced accuracy in identifying a range of emotional expressions, including sadness, joy, surprise, disgust, and fear^26–28. These changes are commonly attributed to age-related cognitive decline affecting neural systems involved in emotional processing, attention, and perceptual integration. Not all studies, however, have identified age-related effects in feline emotion recognition. For example, Nicastro et al.¹² reported no significant association between age and the ability to recognize cat emotions. The small effect size observed in the present study may help reconcile these mixed findings, suggesting that age-related differences, while statistically detectable in large samples, may have limited practical impact.

The present study is the first, to our knowledge, to investigate human ability to interpret cats’ behavioral states using ecologically valid visual stimuli depicting whole-body behavior, including posture, tail position, and facial expressions. Overall, our findings indicate that recognizing feline behavioral states from visual cues alone is a challenging task for humans. Although performance exceeded chance level, accuracy remained relatively low and did not differ robustly across behavioral categories once individual- and stimulus-level variability were taken into account. Rather than being driven by the specific behavioral displayed, performance was shaped primarily by characteristics of the human observer. Gender, age, and ownership emerged as significant predictors of accuracy, highlighting substantial inter-individual variability in the ability to interpret feline emotional-related behavior. These results suggest that humans do not reliably discriminate between different feline behavioral states based solely on visual signals, but instead vary in their overall ability to detect and interpret such cues. At the same time, the observed variability across individual stimulus videos suggests that characteristics of the stimuli themselves—such as behavioral intensity or individual differences in expressivity—may also contribute to recognition performance. Although variability across stimuli was statistically controlled through random effects in the mixed-effects model, future studies would benefit from using a larger and more systematically varied set of video stimuli to directly examine how stimulus-level factors modulate emotion recognition. Moreover, while the use of an online questionnaire enabled the recruitment of a large and diverse sample, participation was voluntary and recruitment occurred via an outreach event and social media. As a result, the sample may reflect a degree of self-selection, potentially favoring individuals with greater interest in cats, as also suggested by the higher proportion of cat owners compared to non-owners in the present sample. Future studies could therefore include specific measures assessing participants’ interest in or attitudes toward cats, as well as additional demographic variables (e.g. ethnicity), in order to directly evaluate how motivational factors influence accuracy in recognizing feline behavioral states. Finally, research conducted in controlled laboratory settings could further clarify how display conditions (e.g. screen size, viewing distance) and stimulus presentation parameters affect the perception of feline emotional-related cues, helping to disentangle perceptual constraints from observer- and stimulus-related sources of variability.

Together, these findings underscore the complexity of human–cat emotional communication emphasizing the pivotal role of individual differences in shaping the perception of feline behavioural state. Importantly, the difficulty in discriminating stress-related states from visual signals alone highlights a potential risk for underestimating early or moderate signs of feline discomfort in everyday interactions. Given the widespread presence of cats in domestic environments, improving people’s understanding of feline behavioural signals is crucial for fostering positive human–cat interactions, reducing misinterpretations, and ultimately promoting animal welfare.

Materials and methods

Participants

A total of 1,950 people of varying genders and ages ranging from 6 to 83 years (32.29 ± 14.99; mean ± S.D.) voluntarily participated to the study. Volunteers were recruited during a science outreach event (European Research Night; N = 634) and via online advertisement on social media (N = 1316).

Stimuli

Participants were shown 12 videos of cats exhibiting three stress-related behavioral states: relaxed, tense, and fearful (Fig. 4). The videos were sourced from a previous study by d’Ingeo et al.⁵, which investigated cats’ emotional responses to human olfactory signals collected in different emotional contexts (fear, happiness, physical stress, and neutral). For each category, four videos were selected, each corresponding to a different cat, resulting in a total of 12 clips (four videos per behavioral state × three behavioral states). A total of 12 cats were included in the present study: they were 8 males and 4 females, aged between 1 and 8 years (4.67 ± 2.65; mean ± S.D.). Video selection was carried out by two veterinary behaviorists with extensive expertise in feline behavior, based on the clearest and most overt visual signals associated with specific behavioral states. The video did not include any audio component; therefore, participants were exposed exclusively to visual information. Behavioral state classification was based on specific behavioral indicators (i.e. body posture, facial expression, tail position) outlined in Table 1, according to the criteria described by Bradshaw³.

Fig. 4.

Representative still frames extracted from the video stimuli used in the study, illustrating cats displaying the three behavioral states: (A) relaxed, (B) tense and (C) fearful.

Table 1.

List of behaviors used to classify the stress-related behavioral state of the cats: relaxed, tense and fearful.

Behavioral state	Behavioral indicators
Relaxed	Relaxed body, eyes, and tail; ears in natural position or held forward.
Tense	Body with tense muscles; eyes wide open or pressed together; ears held forward, sideways, or rotated backward; tail tense with twitching or whipping movements; may blink.
Fearful	Body with tense muscles; wide-open eyes; ears rotated and flattened backward or held forward; tail tense, possibly held close to the body; panting.

Inter-rater reliability was assessed through independent parallel coding of the cats’ behaviours and calculated as percentage agreement, which exceeded 95% for each variable tested. The videos were edited using iMovie to extract short sequences of 3 s. These stimuli were then presented through questionnaires created with Google Forms, where the order of video presentation was pseudo-randomized to ensure that videos showing the same behavioral state were never shown consecutively. All videos featured the cat centrally framed and clearly visible, and were originally recorded in high resolution (4 K), ensuring detailed visualization of body posture, facial features, and tail position.

Procedure

Participants completed the questionnaire individually on their personal devices (i.e. mobile phones). The survey began with a brief demographic section collecting age, gender (female, male, non-binary), and ownership (i.e. “Have you ever had cats?” with “yes” or “no” answers). Participants were then presented with a series of short video clips, each displayed only once. After viewing each video, participants were asked to select the behavioral state they believed the cat was exhibiting, choosing from three options: relaxed, tense, or fearful. Upon selection, the next video was automatically presented. Participants were instructed to watch each video in full-screen mode and not to revise their answers after making a selection, in order to ensure that responses captured their immediate impressions. The test had no fixed duration or time constraints, and participants could complete it at their own pace. The questionnaire link was distributed via social media and through a QR code at the outreach event.

Data analysis

Seventeen participants were excluded due to missing demographic information (gender or age). The initial sample consisted of 1,933 participants (658 males, 1,264 females, and 11 non-binary individuals).

Given the very small number of non-binary participants, these data were not included in the inferential analyses, as reliable estimation of gender-related effects would not have been possible. The final sample therefore comprised 1,922 participants (1,264 females and 658 males), with a mean age of 32.48 years (S.D.= 14.80). Of these participants, 1,249 reported having owned a cat (892 females, 357 males), whereas 673 reported no previous cat ownership (372 females, 301 males).

The data were analysed using a generalized linear mixed-effects model (GLMM) with a binomial distribution and a logit link function, as implemented in the lme4 package in R. Response accuracy (score) was coded as a binary outcome (0 = incorrect, 1 = correct).

Fixed effects included the behavioral state expressed by the cat in the video (relaxed, fear, tense), participant gender (male/female), ownership (yes/no), and age, which was mean-centred prior to analysis. Random intercepts were specified for participants to account for repeated observations, and for stimulus videos to control for variability associated with individual cat stimuli. An interaction between behavioral state and age was initially tested, based on previous evidence indicating an association between age and interspecific emotion recognition^{26,27,29–31}. However, it was not retained in the final model, as it did not improve model fit.

Model parameters were estimated using maximum likelihood estimation with Laplace approximation. Model selection was guided by Akaike (AIC) and Bayesian (BIC) Information Criteria. Statistical significance of fixed effects was assessed using Wald z-tests. When appropriate, regression coefficients were exponentiated to obtain odds ratios (ORs) with corresponding 95% confidence intervals (CIs).

Descriptive accuracy measures by emotion and stimulus video, as well as the confusion matrix, were computed and are reported in the Supplementary Information (Tables S1, S2).

Ethical statement

The study was conducted in accordance with the Declaration of Helsinki (revised in 2013) and European data protection regulations. Participation was voluntary, and the online questionnaire was fully anonymous and did not include any personal information that could identify participants.

All participants were informed about the study’s purpose and procedures prior to participation and provided informed consent. At the beginning of the questionnaire, participants were required to self-declare their eligibility by confirming either that they were 18 years of age or older and consented to participate, or that they were a parent or legal guardian providing consent for the voluntary participation of a minor under their responsibility. Data were processed in aggregate form. The study protocol was submitted to the Ethics Committee of the University of Bari Aldo Moro for evaluation. The Committee determined that, given the anonymous and non-invasive nature of the survey and its compliance with European data protection and privacy regulations, the study did not require formal ethical approval. The research was conducted in accordance with the Committee’s guidance and applicable European data protection directives. Regarding the animal-related component, the study from which the stimuli were derived was approved by the Animal Experimentation Ethics Committee of the Department of Veterinary Medicine (approval no. 19/2020).

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

S.d., M.N., V.S., A.Q. and M.S. designed the research; S.d., M.N., V.S., A.L., A.Q. and M.S. performed the experiments; S.d. and M.S. analyzed the data; S.d., M.N., A.Q. and M.S. wrote the manuscript; S.d., M.N., V.S., A.Q. and M.S. edit the manuscript. All authors reviewed and approved the manuscript.

Funding

This work was supported by Nestlé Purina Petcare, 2020 Sponsorship for Human-Animal Bond Studies to S.d.

Data availability

The dataset analyzed in the current study is included in this published article (as its Supplementary Information).

Declarations

Competing interests

The authors declare no competing interests.