The empirical literature using large international students’ assessments tends to neglect the role of school composition variables in order not to incur in a misidentification of peer effects. However, this leads to an error of higher logical type since the learning environment crucially depends on peers’ family background and on peer heterogeneity. In this paper, using PISA 2006, we show how peer heterogeneity is a key determinant of student attainment and of opportunity equalization. Interestingly, the effect of school compositional variables differs depending on the country tracking policy: peer heterogeneity reduces efficiency in comprehensive systems whereas it has a non-linear impact in early-tracking ones. In turn, linear peer effects are larger in early-tracking systems. Besides, higher heterogeneity tends to equalize student differences related to family background. Results do not change in school- and student-level regressions suggesting that the impact of heterogeneity is correctly identified. Results are also robust when we add school-level dummies and several controls correlated with the school choice to alleviate the selectivity bias of linear peer effects.