This paper makes three contributions to the existing literature. First, it provides descriptive evidence on gender differentials by education level in the US labor market over the last twenty years. Second, it uses the structural estimation of a search model of the labor market to identify and quantify the impact of employers' prejudice on labor market gender differentials. Third, it connects both the descriptive and the analytical findings to recent policy interventions in the US labor market and presents some policy experiments. The results show that prejudice may still have a role in explaining the evidence on gender differentials and there is at least one scenario where the possibility of the presence of prejudiced employers in the labor market has substantial effects. In particular, it is responsible for the reversal of the returns to schooling ranking in recent years and it may explain up to 44% of the gender wage gap of the top education group (Master and PhD) in 2005. Since prejudice is still important, policy interventions may be effective in attaining both efficiency and welfare gains. The paper is in favor of implementing an affirmative action policy because it is frequently able to close the gender gap without reducing overall welfare and because it is effective in targeting the group that should take center stage in the future debate about gender differentials: high-skilled, high-earners workers, who also have family responsibilities.