Statistical discrimination relies on people interpreting information about group differences. Across four pre-registered experiments (n = 7,002), we show that information about the demographics of top performers induces incorrect beliefs that lead to discrimination against minority groups. Participants are less likely to hire women when they receive information about top performers in a male-dominated candidate pool, even when there are no gender differences in performance, and are less likely to hire better-performing non-White candidates when the racial demographics of the candidate pool reflect the US population. A third experiment with neutral group labels shows this is not a motivated error, and a fourth experiment shows that people naturally seek out information about top performers.