Classroom peers are believed to influence learning by teaching each other, and the efficacy of this teaching likely depends on classroom composition in terms of peers' ability. Unfortunately, little is known about peer-to-peer teaching because it is never observed in field studies. Furthermore, identifying how peer-to-peer teaching is affected by ability tracking – grouping students of similar ability – is complicated by the fact that tracking is typically accompanied by changes in curriculum and the instructional behavior of teachers. To fill this gap, we conduct a laboratory experiment in which subjects learn to solve logic problems and examine both the importance of peer-to-peer teaching and the interaction between peer-to-peer teaching and ability tracking. While peer-to-peer teaching improves learning among low-ability subjects, the positive effects are substantially offset by tracking. Tracking reduces the frequency of peer-to-peer teaching, suggesting that low-ability subjects suffer from the absence of high-ability peers to teach them.