This paper develops a model that allows for heterogenous contemporaneous peer effects among different types of agents who are endogenously selected into different peer groups. Using our framework, we characterize the reduced-form coefficient in the peer effect literature and show that it is a priori ambiguous in sign. We apply our approach to migrant and local students in Shanghai, where local students all go to public schools, but migrant students are endogenously selected into either public schools or lower-quality private schools. The results suggest large contemporaneous peer effects among all student groups. We conduct policy experiments to examine the effect of transferring migrant students from private schools to public schools. We show that peer effect can be substantially more important than the school effect in accounting for the total treatment effect of moving to better schools.