We estimate peer effects in cognitive, non-cognitive and health-related outcomes in a sample of Uruguayan high-school students. Our identification strategy is based on two features of the data: (i) parents are not able to choose their child’s class within the school of their choice, and (ii) we observe outcomes over two points in time. We use school and grade fixed effects to avoid confounding peer influence with selection and exploit the longitudinal nature of the data to address Manski’s (1993) reflection problem. We find small but statistically significant peer effects in academic grades (cognitive ability), in having shoplifted, being involved in a fight, and self-satisfaction (proxies for non-cognitive abilities) and in symptoms of depression. The size of the peer effects in non-cognitive and cognitive skills are of similar magnitude. We also find that popular students are less likely to be influenced by peers in terms of cognitive outcomes and that peer effects in non-normative behavior (shoplifting) and depression are milder for students with higher levels of household education.