Although there is debate about whether juvenile incarceration deters future crime, it is a common practice worldwide. We contribute to this debate by using Chilean data to assess the causal impact of different types of juvenile incarceration on recidivism in young adulthood (18-21 years old). To address the endogeneity issues, we use the quasi random assignment of detention judges as instrumental variable to estimate the effect of pretrial detention, and the quasi random as- signment of public attorneys to estimate the effect of any type of incarceration. Considering a standard IV linear model, we find that pretrial detention increases the probability of recidivism by 61 percentage points (pp), and when we define the treatment as any type of incarceration, this impact is equal to 65 pp. When we estimate bivariate probit models – using a novel approach for estimating this model in the context of fixed effects – the impact of pretrial detention and incar- ceration on recidivism are equal to 12 pp and 15 pp, respectively. We also estimate the marginal treatment effect (MTE), finding that the magnitudes of the marginal effects are larger for those individuals with low treatment probabilities. If we use MTE estimates to calculate the average treatment effect (ATE), the impact of pretrial detention on recidivism is equal to 28 pp. If we define the treatment as any type of incarceration, this impact is equal to 36 pp. Finally, we find that an important mechanism behind these impacts is the effect of these different types of incarceration on high school graduation.