We examine the effect of peer achievement on students' own achievement and teacher performance in primary schools in disadvantaged neighborhoods using data from a well-executed randomized experiment in seven states. Contrary to the existing literature, we find that the average classroom peer achievement adversely influences own student achievement in math and reading in linear-in-means models. Extending our analysis to take into account the potential non-linearity in the peer effects leads to non-negligible differences along the achievement distribution. We test several models of peer effects to further understand their underlying mechanisms. While we find no evidence to support the monotonicity model and little evidence in favor of the ability grouping model, we find stronger evidence to support the frame of reference and the invidious comparison models. Moreover, we also find that higher achieving classes improve teaching performance in math. Finally, using a simple policy experiment we find suggestive evidence that tracking students by ability potentially benefits students who end up in a low achievement class while hurting students in a high achievement class.