We investigate possible racial discrimination in the context of discretionary parole release. We develop a rational choice model of release whereby a parole board must balance parolees' risk of violation with the cost of not releasing prisoners who may not violate their parole. A color-blind parole board would release all individuals below a certain risk threshold. To test this prediction, we take advantage of a unique data set that reports all prisoners released on parole between 1983 and 2003 in the U.S. We apply the outcome test methodology recently used to assess racial profiling in police search decisions. Here, a higher rate of parole violation within a group suggests that the parole board used a less restrictive paroling criterion, and is thus biased in favor of that group. To overcome the concern of inframarginality that traditionally plagues outcome tests we provide evidence that parole boards strategically time the release of parolees. In turn, both minority and White prisoners become marginal from the perspective of their probability of parole violation. Parole boards operating under an indeterminate sentencing regime appear biased against White prisoners whose violation rate is significantly smaller than that of African Americans. In contrast, this gap is smaller or null when there is no discretion in the paroling system. Further evidence rules out post-release discrimination. We propose different hypotheses to account for the evidence.