The Eyes Say It Differently: Brazilian Study Reveals Cultural Bias in a Famous Empathy Test

For more than two decades, one of psychology’s most beloved instruments has asked people to do something deceptively simple: look at a black-and-white photograph of a stranger’s eyes and guess what that person is feeling. The Reading the Mind in the Eyes Test, developed by Simon Baron-Cohen and colleagues in 2001, has been administered to millions of participants across dozens of countries, used to probe empathy in autism, schizophrenia, depression, bipolar disorder, and countless other conditions. But a new study from Brazilian researchers suggests that one of the test’s most fundamental assumptions may not travel well across cultures. When the researchers asked Brazilian adults to rate the emotional tone of the test’s items, they found a pattern of emotional valence that diverged markedly from the pattern reported in a comparable English-speaking sample, raising uncomfortable questions about how much of what we call mind-reading is actually culture-reading.

The study, published in the journal Trends in Psychology, was led by Laura Canani and Letícia S. Czepielewski of the Cognition and Psychopathology Laboratory at the Federal University of Rio Grande do Sul, working with Sarah Corrêa de Sales and Raffael Massuda of the Psychosis Research Program at the Federal University of Paraná. Their goal was not to question the test’s basic utility but to fill in a technical gap that has long been acknowledged by specialists: although the Reading the Mind in the Eyes Test is routinely described as a measure of decoding complex mental states, the instrument itself does not classify the emotional valences of the mental states it presents. In other words, the test tells you whether someone can pick the right label for a pair of eyes, but it says nothing systematic about whether that label describes something pleasant, unpleasant, or emotionally neutral.

That gap matters more than it might first appear. Emotional valence, the degree to which a mental state is experienced as positive or negative, is one of the organizing dimensions of affective science, and it shapes how people attend to, remember, and interpret social information. Clinicians and researchers have long suspected that patients with depression, for example, may show biased processing of negative emotional material, and that such biases could interact with performance on mental-state decoding tasks. Without knowing the valence of each item in the test, however, researchers cannot cleanly separate the ability to decode a mental state from the tendency to be drawn to, or repelled by, emotionally charged material. A person who performs poorly on the test might genuinely struggle to read minds, or might simply avoid guessing negative states, and the standard scoring system cannot distinguish between these possibilities.

The Brazilian team followed a methodological template laid out by Catherine Hudson and colleagues in a 2020 study published in Psychological Assessment, which was the first to systematically classify the valence of the test’s items using an English-speaking sample. Canani and her colleagues recruited 186 Brazilian adults and collected data through an online cross-sectional quantitative survey. Participants were asked to evaluate the emotional character of the words and picture-word pairs that make up the revised version of the test, which contains 36 items pairing photographs of the eye region with four candidate mental-state labels, and 45 items consisting of the mental-state words alone. The researchers used the Brazilian version of the test, which had previously been adapted and validated for Portuguese speakers, ensuring that language barriers would not contaminate the valence judgments.

The results were striking. Among the 36 picture-word pairs, participants classified 7 items as neutral, 12 as positive, and 17 as negative. Among the 45 words-only items, the distribution tilted even more heavily toward the unpleasant end of the spectrum: 13 words were judged positive, only 1 was judged neutral, and 31 were judged negative. This heavy skew toward negative valence in the vocabulary of the test is itself revealing. The mental-state words that Baron-Cohen and his team selected to capture the subtleties of human inner life, words like skeptical, uneasy, or defensive, are simply not emotionally balanced, and Brazilian raters perceived that imbalance clearly.

More consequential still was the comparison with the earlier English-language study. When the Brazilian classification was set against the valence assignments produced by Hudson and colleagues, important differences emerged, suggesting that the same mental-state vocabulary does not carry the same emotional weight for speakers of different languages embedded in different cultural worlds. The authors interpret this divergence through the lens of cultural models of emotion, a framework in which the meanings, norms, and display rules that surround emotional experience are shaped by the societies in which people live. Classic work in cultural psychology, including the influential theory of independent and interdependent self-construals proposed by Hazel Markus and Shinobu Kitayama, has long argued that cognition, emotion, and motivation are inseparable from cultural context. The new findings extend that argument into the fine-grained territory of psychometric measurement, showing that even the perceived emotional tone of a single word can shift across cultural boundaries.

The implications ripple outward across an enormous research literature. The Reading the Mind in the Eyes Test has been used to document social cognitive dysfunction as a clinical marker across some 30 psychiatric and neurological conditions, from borderline personality disorder to post-traumatic stress disorder, from first-episode psychosis to eating disorders. Large-scale cross-cultural studies, including a 2022 analysis of sex and age differences in theory of mind across 57 countries, have relied on the English version of the test to draw sweeping conclusions about how mentalizing varies around the globe. If the emotional valence of the test’s items is culturally contingent, then some portion of the cross-cultural variance in test scores could reflect differences in how emotional words are evaluated rather than genuine differences in the capacity to infer mental states. A Brazilian participant and a British participant might both correctly identify the mental state behind a pair of eyes, yet weight its pleasantness differently, and downstream analyses that pool valence information across samples could be quietly confounded.

The study also speaks to a broader movement in psychiatry toward frameworks such as the Research Domain Criteria, which treat social cognition as a core domain of functioning and call for measures that can cleanly parse its subcomponents. Within that vision, knowing the valence structure of a widely used task is not a pedantic detail but a prerequisite for building theoretically coherent models of how emotion recognition breaks down in psychopathology. The Brazilian authors note that impairments in social cognition have been documented across a wide range of conditions and lead to serious difficulties in social functioning, which makes it all the more important that the instruments used to detect those impairments behave consistently and transparently. A test whose items carry culturally variable emotional charges cannot fully serve that purpose without supplementary information about what each item means emotionally to the population being assessed.

Technically, the study demonstrates a practical and relatively inexpensive pathway for researchers elsewhere: recruit a reasonably sized community sample, present the test’s items for valence judgment, and compare the resulting classification against existing benchmarks. The Brazilian team’s use of an online cross-sectional design, combined with the previously validated Portuguese adaptation of the test, shows that valence mapping can be carried out without laboratory equipment or clinical populations, making it feasible for research groups in many countries. The authors analyzed their data using standard quantitative methods implemented in the R statistical environment, and the work was conducted in accordance with Brazilian research ethics regulations, including the national data protection law governing the handling of personal information.

What emerges from this study is a picture of a famous psychological instrument that is both more and less universal than its reputation suggests. The test’s photographs of eyes may tap into something deeply human about the way we search faces for meaning, but the emotional labels attached to those photographs are filtered through language, culture, and lived experience. As the authors conclude, culture models appear to play an important role in how emotional valence is classified, and researchers who use the Reading the Mind in the Eyes Test outside the English-speaking world would be wise to establish the valence profile of the items for their own populations. In the science of reading minds, it turns out, the first thing we need to read more carefully is our own instruments.

Subject of Research: Cultural differences in the emotional valence of Reading the Mind in the Eyes Test items among Brazilian adults

Article Title: Emotional Valences in the Reading the Mind in the Eyes Test: A Brazilian Study

Article References: Canani, L., Corrêa de Sales, S., Massuda, R., & Czepielewski, L. S. (2025). Emotional Valences in the Reading the Mind in the Eyes Test: A Brazilian Study. Trends in Psychology. https://doi.org/10.1007/s43076-025-00503-6

Image Credits: AI Generated

DOI: 10.1007/s43076-025-00503-6

Keywords: social cognition, theory of mind, Reading the Mind in the Eyes Test, emotional valence, emotion recognition, cultural psychology, psychometrics, psychological assessment, Brazil, mental state decoding, empathy, psychopathology