A growing literature emphasizes the importance of social skills in the labor market. However, to date, no study addresses the role of peer characteristics in the formation of social skills. This paper reports estimates of cognitive and social peer effects from a large-scale field experiment at selective boarding schools in Peru. My experimental design overcomes some methodological challenges in the peer effects literature. I randomly varied the characteristics of neighbors in dormitories with two treatments: (a) less or more sociable peers (identified by their position in the school’s friendship network before the intervention) and (b) loweror higher-achieving peers (identified by admission test scores). While more sociable peers enhance the formation of social skills, higher-achieving peers do not improve academic achievement; in fact, they further reduce the academic performance of lower-achieving students. These results appear to be driven by students’ self-confidence and the social support they receive from their neighbors. I interpret these findings in the context of a simple self-confidence model where students infer their skills by interacting with their peers.