Ability grouping is a controversial issue in education (Slavin, 1987). Proponents argue that it facilitates teaching and learning, opponents criticise the lack of a positive reference group for low achievers as well as social and ethnic disparities (Oakes, 1985). International research findings are inconclusive, but show mostly weak effects. For the German context, Lehmann (2006) finds a strong advantage for heterogeneous classes. Analyses based on the longitudinal study KESS in Hamburg show small advantages of heterogeneous classes for the development of reading achievement in secondary school, especially for low achievers (Bos & Scharenberg, 2010; Scharenberg, 2012). We take up these analyses and focus on three key questions with regard to Mathematics achievement: Do primary school students learn more in homogeneous or heterogeneous classes? To what extent is heterogeneity confounded with other compositional effects? Can high- and low-ability students equally benefit? Drawing on data from the ADDITION project, multi-level analyses control for gender, cultural and social background and prior knowledge on the student level (n= 1.119) as well as aggregated variables on the class level. According to our previous research we assume that primary school students can benefit from heterogeneous classes. Low-ability students are expected to profit most.