We study sources of racial di fferences in criminal justice outcomes using administrative data from Texas, North Carolina, and Arkansas. We document large racial disparities in incarceration conditional on rich controls for case and defendant characteristics. Black defendants are 20-46% more likely to receive an incarceration sentence than similar white defendants arrested for the same off ense. Disparities emerge in both charging and sentencing. In Texas and Arkansas, over 40% of the unexplained racial gap is between-jurisdiction; black defendants are concentrated in counties where courts treat all defendants more punitively. Exploiting within- defendant variation in case jurisdiction, we argue that the across-jurisdiction variation in punishment we identify reflects causal diff erences across court systems in punitiveness.  Moreover, we argue that the relationship between jurisdiction punitiveness and the black share of the arrested can be explained by in-group bias-voters prefer more severe punishment when off enders are more likely to belong to a different racial group. This mechanism suggests that the relationship will follow an inverted U-shape; while white voters prefer more punitive policy as the black share of defendants increases, for jurisdictions with sufficiently large black populations, the pivotal voter is more likely to be black. This prediction is borne out in the
data. Data on racial animus and policy preferences by race and local demographics provide additional support. We conclude by simulating racial confi nement gaps under a counterfactual in which all jurisdictions adopt the punishment level imposed by racially homogeneous jurisdictions. Under this counterfactual, both the absolute confi nement rate gap and the overall confi nement rate fall by over 50%.