We exploit rules of class formation to identify and estimate the causal effect of increasing the number of immigrants in a classroom keeping class size constant. Since principals use class size to neutralize the possible detrimental effects of immigrants' inflows, this composition effect is the relevant policy parameter but it has been neglected by the literature so far. Our empirical analysis is guided by a model of the educational production function that allows for two types of students, natives and immigrants, who have a non-negative probability of generating disruption or positive externalities across or within groups. We show that the pure composition effect is negative both in language and in mathematics.