We employ a standard identification strategy from the peer effects literature to investigate the importance of network definitions in estimation of endogenous peer effects. We use detailed information on friends in the Adolescent Longitudinal Health Survey (Add Health) to construct two network definitions that are less ad hoc than the school-grade cohorts commonly used in the educational peer effects literature. We demonstrate that accurate definitions of the peer network seriously impact estimation of peer effects. In particular, we show that peer effects estimates on educational achievement, smoking, sexual behavior, and drinking are substantially larger with our more detailed measures than with the school-grade cohorts. These results highlight the need to further understand how friendships form in order to fully understand implications for policy that alters the peer group mix at the classroom or cohort level.