We study the effects of grade retention on secondary school performance by considering a change in Colombia's educative legislation. In 2010, the rule that forced schools to retain up to a 5% of students was abolished. Exploiting variation in schools' retention rates in a difference-in-differences framework, we find that retained (marginally non-retained) students improve (decline) their performance on language but not on math test scores. We suggest the school's position in the retention distribution, and the proportion of inexperienced teachers in the classroom, can be the mechanisms by which the marginally decreasing returns of grade retention are determined.