We study an experimental market where, in spite of equal performance across genders, individuals discriminate towards women in hiring decisions. We show that discrimination is neither taste-based nor based on a correct statistical inference regarding differences in performance. Instead, it is rooted in biased beliefs about women’s abilities. The gender gap increases when candidates are allowed to influence expectations by declaring their expected performance and it narrows if individuals receive accurate information of the performance of the applicants. However, even when accurate information is transmitted, the gender gap persists because individuals do not completely update their initially biased belief. Furthermore, we show, by using the Implicit Association Test, that unconscious stereotypes are partly responsible for the initial bias in beliefs and the subsequent lack of updating.