In Trinidad and Tobago students are assigned to secondary schools after fifth grade based on achievement tests, leading to large differences in the school environments to which students of differing initial levels of achievement are exposed. Using both a regression discontinuity design and rule-based instrumental variables to address self-selection bias, I find that being assigned to a school with higher-achieving peers has large positive effects on examination performance. These effects are about twice as large for girls than for boys. This suggests that ability-grouping reinforces achievement differences by assigning the weakest students to schools that provide the least value-added.