This research examined bias and accuracy in judging hostile and benevolent sexism during mixed-gender interactions. Bias is defined as underestimation or overestimation of a partner’s sexism. Accuracy is defined as covariation in two different ways, as (a) the strength of the association between a dyad member’s judgment and their partner’s sexism, across dyads, and (b) the ability to differentiate sexism between multiple targets. In Studies 1 and 2, members of mixed-gender dyads rated their own and their partners’ benevolent and hostile sexism. Overall, there was no covariation, across dyads, between perceptions and the partner’s self-reported sexism. However, women overestimated men’s hostile sexism; there was no evidence of biases for women judging men’s benevolent sexism. Men underestimated women’s hostile sexism and overestimated benevolent sexism. In Study 3, participants watched brief videos of male or female students (targets) from Study 1 and completed benevolent or hostile sexism items for each target as they thought the target would fill them out. Accuracy for detecting sexism across multiple targets (using sensitivity correlations) was significantly above chance for both forms of sexism. Male and female participants were more accurate at detecting hostile sexism in male targets than female targets. Participants were more accurate at detecting benevolent sexism of same-gender targets than opposite-gender targets. When judging targets of opposite gender, women’s judgments were significantly above chance for both forms of sexism, but men were not accurate for either forms of sexism. These studies suggest that there is bias and accuracy in first impression judgments of sexism.