This paper estimates the treatment effect of grade repetitions in French junior high schools, using a value-added test score as outcome and quarter of birth as instrument. With linear two-stage least squares, local average treatment effect is estimated at around 1.6 times the standard deviation of the achievement gain. With non-linear full-information maximum likelihood, average treatment on the treated ranges between one and one-quarter of the standard deviation of the outcome. Grade repetition in junior high school is also shown to increase the probability of graduating from junior high school by 2.5 probability points.