Cross-Race Effect: Why Other-Group Faces Are Harder to Tell Apart

The effect is a difference in discrimination, not in attention or goodwill. When people study faces and later sort old from new, own-group faces produce more hits and fewer false alarms than other-group faces; Meissner and Brigham called this a mirror effect, because the two kinds of error move in opposite directions at once. Other-group faces are both harder to remember having seen and easier to mistake for ones you have seen. The practical consequence in a lineup is that a witness is more likely both to miss the true culprit and to pick an innocent lookalike.

Two families of explanation have the most support, and current models combine them. The perceptual expertise account holds that people learn, through years of exposure, to encode the features that distinguish the faces they see most, and that this tuning does not transfer to a group whose faces vary along somewhat different dimensions. On this view the effect is a skill gap, and contact should narrow it; the meta-analysis found that it does, modestly, with contact accounting for about 2 percent of the variability. The social-cognitive account holds that faces from an out-group are processed as members of a category rather than as individuals, so the details that would tell them apart are never encoded. Hugenberg, Young, Bernstein and Sacco's categorization-individuation model of 2010 brings the two together: motivation to individuate determines whether expertise gets applied, and in laboratory studies telling people about the effect and asking them to attend to individual features has reduced it, with mixed results on how reliably.

The effect is a group average with wide individual variation. It appears in the participant groups the studies have compared and is not a measure of attitude: a person can hold no hostility toward a group and still find its faces harder to tell apart, which is what the expertise account predicts. In the meta-analysis the own-group advantage was found across the samples studied, with its size varying by sample and by method. None of this makes any single cross-race identification wrong. It makes a class of identifications less reliable on average, which is a fact about procedures rather than about any witness.

Roy Malpass and Jerome Kravitz published "Recognition for faces of own and other race" in the Journal of Personality and Social Psychology in 1969. They photographed 40 Black and 40 white college-age men under conditions chosen to reduce cues other than the face itself. Participants were 26 undergraduates at the University of Illinois, 13 Black and 13 white, and 14 first-year graduate students at Howard University, 7 Black and 7 white. Each saw 20 slides drawn from the pool of 80, then all 80, and marked the ones they had seen. White faces were recognized better overall in both samples, and the interaction between the race of the participant and the race of the face, the signature of an own-race advantage, was found in the Illinois data and marginally in the Howard data. The paper established the paradigm that the next three decades of work would use: study a set of faces, then sort old from new.

Christian Meissner and John Brigham's 2001 meta-analysis in Psychology, Public Policy, and Law, "Thirty years of investigating the own-race bias in memory for faces," is the reference the demo quotes. It combined 39 articles, 91 independent samples and nearly 5,000 participants, examined hit rates, false alarm rates, discrimination accuracy and response criterion, and tested eight moderators along with the roles of racial attitudes and interracial contact. The results showed the mirror pattern, with own-race faces yielding more hits and fewer false alarms, participants 1.40 times more likely to correctly recognize an own-race face they had seen and 1.56 times more likely to falsely recognize a novel other-race face, and interracial contact playing a small but reliable role, accounting for about 2 percent of the variability across participants.

Later reviews have taken up the theory and the application. Young, Hugenberg, Bernstein and Sacco's 2012 review in Personality and Social Psychology Review weighed the perceptual expertise and social-cognitive accounts and argued for hybrid models. Wilson, Hugenberg and Bernstein's 2013 article in Social Issues and Policy Review set out what the effect means for eyewitness procedure.

The rows below are illustrations, not findings from the studies cited on this page.

The effect is a limit on a perceptual skill, so the counters are procedures that stop a single face memory from carrying more weight than it can bear, plus habits that build the skill.

• Treat confidence as one signal, not the signal. Cross-race identifications carry more false alarms on average, and the evidence on accuracy at matched confidence is mixed: a 2017 reanalysis found no reliable difference above chance, a 2025 lineup study found high-confidence cross-race accuracy trending lower. Confidence alone neither discredits a cross-race identification nor confirms it.
• Fix procedures around the limit. Double-blind lineups, a record of the witness's confidence at the moment of the first identification, and a search for independent corroboration protect against a wrong pick regardless of its cause.
• Individuate on purpose. In laboratory studies, telling people about the effect and asking them to attend to what makes each face distinct has reduced it, though not consistently across studies; the habit costs nothing and is the one lever an individual has in the moment.
• Build experience with individual faces. Contact narrows the gap modestly in the research, and the expertise account suggests that what matters is learning to tell individuals apart, not simply being around a group.
• Expect the effect in yourself. The demo's second question is the one to ask before trusting a cross-group face memory: would I do as well on these faces as on familiar ones? The research says most people would not.

The cross-race effect is a memory finding about faces. Its neighbors are judgments about people.

• It is not the out-group homogeneity bias, the belief that members of other groups are more alike in their traits and opinions than members of your own. That is a judgment about personalities; the cross-race effect is a measured difference in recognizing faces, and it can occur without any belief about the group at all.
• It is not prejudice, an attitude toward a group. The effect is found regardless of what participants believe, and the meta-analysis found only a small relation between contact and its size; attitudes and face memory are different measures.
• It is not stereotyping, assigning traits to a person from their group. Stereotyping fills in what a person is like; the cross-race effect concerns whether you can tell which person it is.
• It is not implicit stereotypes, automatic associations between a group and attributes, although the categorization-individuation model places both downstream of the same step: processing a face as a category member first.