Using exogenous secondary school assignments to remove self-selection bias to schools and peers, I obtain credible estimates of (1) the effect of attending schools with higher-achieving peers, and (2) the direct effect of peer quality improvements within schools, on the same population. While students at schools with higher-achieving peers have better academic achievement, within-school increases in peer achievement improve outcomes only at high-achievement schools. Estimates suggest that peer quality can account for over half of school value-added among the top quartile of schools, but little value-added for other schools. The results reveal some large and important differences by gender.