Recidivism rates are a growing concern due to the high cost of imprisonment and the high rate of ex-prisoners returning back to prison. The factors leading to recidivism are multifaceted, but one policy-relevant and potentially important contributor is the composition of peer inmates. In this paper, we study the role of peer effects within a correctional facility using data on almost 80,000 individuals serving time in Georgia. We exploit randomness in peer-composition over time within prisons to identify effects of peers on recidivism rates. We find no evidence of peer effects for property and drug-related crimes in the general prison population. However, we find strong peer effects when we define peer groups by race and age. Our findings indicate that homophily plays a large part in determining the strength of peer exposure among prisoners in the same facility. Our findings suggest that prison assignments can be a way to reduce recidivism for particular groups of prisoners.