This paper studies how the presence of peers and different peer assignment rules—self-selection versus random assignment—affect individual performance. Using a framed field experiment, we find that the presence of a randomly assigned peer improves performance by 28% of a standard deviation (SD), whereas self-selecting peers induces an additional 15%–18% SD improvement in performance. Our results document peer effects in multiple characteristics and show that self-selection changes these characteristics. However, a decomposition reveals that variations in the peer composition contribute only little to the performance differences across peer assignment rules. Rather, we find that self-selection has a direct effect on performance. This paper was accepted by Yan Chen, decision analysis.