Children routinely benefit from being assigned a teacher who shares an identity with them, such as gender or ethnicity. We study how student beliefs impact teacher-student gender match effects, and how this varies across subjects with different societal beliefs about differential ability by gender. A simple model of belief formation predicts that match effects will be larger for students who believe they are of low ability, and be greater in subjects with more salient societal beliefs. We test these using data from Chinese middle schools, exploiting random assignment of students to teachers. In China, many people believe boys are innately better than girls at math. We find that being assigned a female math teacher helps low-perceived-ability girls and slightly harms low-perceived-ability boys, with no effects for other children. In English and Chinese – subjects with less salient societal beliefs – these patterns persist but diminish. This yields policy implications for the assignment of teachers to students.