A growing number of American states require that students who do not demonstrate basic reading proficiency at the end of third grade be retained and provided with remedial services. We exploit a discontinuity in the probability of third grade retention under Florida's test-based promotion policy to study the causal effect of retention on student outcomes over time. Although conventional OLS estimates suggest negative effects of retention on achievement, regression discontinuity estimates indicate large positive effects on achievement and a reduced probability of retention in subsequent years. The achievement gains from test-based retention fade out over time, however, and are statistically insignificant after six years.