This paper investigates whether people have distorted beliefs about the ability of the social groups they identify with. We find that experimentally manipulated identification with a randomly composed group leads to overconfident beliefs about fellow group members' performance on an intelligence test. We also find that stronger group identification leads people to put more weight on positive signals, and less weight on negative signals, about their group's relative performance on the test. Our results suggest that beliefs about group ability can be distorted in a similar fashion as beliefs about individual ability. This in-group bias in beliefs has important economic consequences when group membership is used to make inference about an individual's characteristics, as in hiring or judicial decisions. We discuss the potential implications of this bias for economic theories of discrimination.