Abstract We explore how feedback shapes, and perpetuates, gender gaps in self-assessments. Participants in our experiment take tests of their ability across different domains. We elicit their beliefs of their performance before and after feedback. We find that, even after the provision of highly informative feedback, gender stereotypes influence posterior beliefs, beyond what a Bayesian model would predict. This is primarily because both men and women update their beliefs more positively in response to good news when it arrives in a more gender congruent domain (i.e. more male-typed domains for men, more female-typed domains for women), fueling persistence in gender gaps.