Abstract In this laboratory experiment, we show that people incorporate irrelevant group information when evaluating others. Individuals from groups that perform badly on average receive low evaluations, even when it is known that the individuals themselves perform well. This group-bias occurs both in a gendered setup, where women form the worse performing group, and in a non-gendered setup. The type of discrimination that we identify is neither taste-based nor statistical; it is rather due to conservatism in updating beliefs, and is even more pronounced among women. Furthermore, self-confident men overvalue male performers. When our data is used to simulate a job promotion ladder, we observe that women are driven out quickly.