Using Dutch data on pupils in elementary school this paper is the first empirical study to analyze whether assigning homework has a heterogeneous impact on pupil achievement. Addressing potential biases by using a difference-in-difference approach, I find that the test score gap is larger in classes where everybody gets homework than in classes where nobody gets homework. More precisely pupils belonging to the upper part of the socioeconomic scale perform better when homework is given, whereas pupils from the lowest part are unaffected. At the same time more disadvantaged children get less help from their parents with their homework. Homework can therefore amplify existing inequalities through complementarities with home inputs. (Contains 1 figure and 11 tables.)