Using detailed administrative data and quasi-random cohort-level variation, I find that exposure to high risk peers while in a juvenile correctional facility has a large impact on future crime. I consider three mechanisms to explain this effect: criminal skill transfer, the formation of criminal networks which persist after release, and the social contagion of crime-oriented attitudes and non-cognitive traits. I find evidence consistent with the social contagion mechanism in residential correctional facilities. Exposure to peers from unstable and/or abusive homes leads to increased aggression, impulsivity and anti-societal attitudes, as well as increased criminal activity. Geographical constraints make persistent network formation unlikely in this setting, and the effect is not diminished by controlling for criminal experience (a proxy for skill). I find evidence consistent with persistent network formation in alternative schools for delinquent youth, a more conducive setting for this mechanism due to geographical concentration. Exposure to peers with extensive criminal experience or gang connections influences future criminal behavior, but only if peers live in the same city. I examine differences across genders: while boys are particularly susceptible to negative peer influence, girls respond more to the positive influence of lower risk peers. Using a variety of tests I find no evidence of correlation in criminal propensity across cohorts once facility and time-of-release fixed effects have been accounted for. I develop a "split sample" technique to generate unbiased estimates of the correlation between own and peer risk levels despite negative mechanical correlation, and verify standard errors non-parametrically via permutation.