We study the transmission of beliefs about gender differences in math ability from adults to children and how this affects girls' academic performance relative to boys. We exploit randomly assigned variation in the proportion of a child's middle school classmates whose parents believe boys are innately better than girls at learning math. An increase in exposure to peers whose parents report this belief increases a child's likelihood of believing it, with similar effects for boys and girls and greater effects from peers of the same gender. This exposure also affects children's perceived difficulty of math, aspirations, and math performance, generating gains for boys and losses for girls.