We use administrative panel data on about a quarter of a million students in the German state of Hesse to estimate the causal effect of class size on school tracking outcomes after elementary school. Our identification strategy relies on the quasi-random assignment of students to different class...