This paper uses administrative longitudinal micro data on the universe of Junior High school students in Uruguay to measure the effect of grade failure on students' subsequent school outcomes. Exploiting the discontinuity induced by a rule establishing automatic grade failure for pupils missing more than 25 days, I show that grade failure leads to substantial drop-out and lower educational attainment even after 4 to 5 years since the time when failure first occurred. Complementary evidence based on a change in the regime of grade promotion leads to very similar conclusions, suggesting that non-random sorting around the discontinuity point is unlikely to drive my results.