Peer effects are an important contributing factor in the learning process. Most of the prior literature on peer effects focuses on the characteristics of peers rather than examining the structure of peer networks. We attempt to measure not only the impact of peers but also the structure of the peer network. In particular we are interested in the characteristics of students' study groups along several dimensions: quality, heterogeneity, size and cohesion. Using pre-college characteristics of students and a random assignment into sections in their first year, we construct instruments of the study group measures to control for endogeneity of the network formation. Our OLS and IV estimates suggest that peer quality improves student performance, and that the breadth and cohesion of students' network positively affects student outcomes. We also find significant heterogeneity of the results depending on network characteristics. Our findings can be used to assist university administrators or professors to choose criteria for sorting students into study groups.