Military coups are often advocated as solutions for ending state-sponsored atrocities. Yet, we know little about coups’ precise consequences. This article estimates the effect of coups on state repression by exploiting the element of chance in whether an attempted coup succeeds or fails. Contrary to popular views of coups as remedies against repressive autocrats, I find no evidence that coups have a pacifying effect on state repression. Rather, coups appear to make matters worse, even when targeting leaders who commit large-scale human rights violations. This article contributes to studies of political violence, authoritarianism, and civil-military relations by resolving a long-standing “good coup versus bad coup” debate. It also advances literature on coups and their consequences through an innovative empirical design that leverages exogenous variation in coup outcomes, combined with an extreme bounds analysis, overcoming conventional challenges of causal inference using observational data.