Abstract: Using two surveys, we study how respondents process visual cues to identify the political orientation (left- vs. right-wing) of French deputies, based on their official photographs only, to test the type of heuristic that they use. We first confirm that respondents outperform random guesses. Second, we find that their categorizations correlate with observable characteristics (gender, color of the tie, jewelry) and subjective assessments of deputies’ personality traits (attractiveness, competence, trustworthiness). Third, the cues that respondents use are consistent with the actual characteristics of left- and right-wing deputies. Fourth, the magnitude of the marginal impact of a characteristic on the probability that a respondent categorizes a photograph as left- or right-wing increases strictly with the representativeness of that characteristic. Finally, we find evidence that some characteristics correlate with categorization errors. Those findings are at odds with Bayesian inference but consistent with the representativeness heuristics suggested by Kahneman and Tversky (1972) and recently modelled by Bordalo, Gennaioli, and Shleifer (2014).