This paper examines peer effects in a Chinese middle school where: 1. classes are randomly assigned to teachers, and 2. student quality across classes varies because student assignment is based on a noisy measure of student quality. Peer effects are concentrated primarily on math scores, as opposed to Chinese or English scores. Improvements in peers at the bottom of the distribution of ability leads to improvements in student performance both for the student’s own class and for sibling classes that share the same teachers, but are not connected in any other way. For middle and top tercile peers, improvements in peers appear to reduce student test scores. The positive effects of peers at the bottom of the distribution are primarily associated with the ability of boys in the class and the sibling class, while the negative effect of peers seems to be driven by same gender peers. Finally, the positive own class and sibling class peer effects arise primarily when the head teacher of the class or the sibling class, respectively, teaches math.