Using unique geo-coded information on the residential address of a representative sample of American adolescents and their friends, we revisit the importance of geographical proximity in shaping education outcomes. Our findings reveal no evidence of residential neighborhood effects. Social proximity, as measured by similarity in religion, race and family income as well as in unobserved characteristics, appears to play a major role in facilitating peer influence. Our empirical strategy is able to control for the endogeneity of both social network and location choices.