Background: The effects of school socio-economic composition on student achievement growth trajectories have been a hot topic of discussion among politicians around the world for many years. However, the bulk of research investigating school socio-economic composition effects has been limited in important ways. Aims: In an attempt to overcome the flaws in earlier research on school socio-economic composition effects, this study used data from a large sample, followed students throughout primary education, addressed selection bias problems, identified the grade(s) in which school socio-economic composition mattered the most, and studied the differential effects of school socio-economic composition by individual socio-economic status (SES). Sample: In a longitudinal design with seven occasions of data collection, the authors drew on a sample of N = 3,619 students (age at T1 about 5 years, age at T7 about 12 years) from 151 primary schools in Flanders (the northern part of Belgium). Method: Students in low-, medium-, high-, and mixed-SES schools were matched using propensity scores. To compare students' achievement growth trajectories in the different school compositions, multilevel regression modelling with repeated measurements was applied. Results: The results showed that students had more positive achievement growth in high-SES as compared to low-SES and mixed-SES schools. In two of the three comparisons, students in mixed-SES schools showed the lowest math development. The negative effects of mixed-SES schools on math achievement growth were the strongest for high-SES students. Conclusions: Our findings contribute to the ongoing discussion on the effects of school socio-economic composition on student achievement growth.