This article uses the nationally representative State Court Processing Statistics on felony defendants to analyze how judges decide if defendants should be held pretrial. We find a large (11 percentage points) racial gap in hold rates within a county. Judicial decisions are significantly influenced by the probability that the defendant will be rearrested pretrial for a violent felony. Controlling for this probability causes the racial gap in hold rates to disappear. Bail amounts follow the same pattern. The most plausible sources of bias—mismeasurement of the control probabilities or selection bias—likely either do not matter or cause an upward bias.