We examine the impact of language training on the economic integration of immigrants in France. The assignment to this training, offered by the French Ministry of the Interior, depends mainly on a precise rule: the training is provided if the test score of an initial language exam is below a certain threshold. This eligibility rule creates a discontinuity in the relation between the test result and the variables of interest, which is used to estimate the causal effect of language training, through the method of Regression Discontinuity Design. We find that the number of assigned hours of training significantly increases labor force participation of the treated individuals. The language classes appear to have a larger effect for labor migrants and refugees relative to family migrants, for men and individuals below the median age, and for individuals with higher levels of education. Our estimated coefficients are remarkably similar when we rely on local linear regressions using the optimal bandwidth with few observations around the threshold and when we control parametrically for a polynomial of the forcing variable and use the whole estimation sample. We discuss extensively why manipulation of the entry test score is theoretically unlikely and show robustness checks that consider the possibility of misclassification. We conclude with a discussion of the candidate mechanisms for the improved labor market participation of immigrants.