We use linked employer-employee data to investigate a large sample of past and future prisoners in Hungary, 2003-2011. We first compare their jobs, focusing on attributes that can reduce the penalty the employer must pay for a mistaken hiring decision. Second, we study if employers insure themselves by paying lower wages to ex-prisoners. Third, we analyze whether the probability of the match dissolving within a few months is lower if the firm could potentially base its hiring decision on referrals. The composition of former prisoners' employment is biased toward easy-to-cancel jobs. In the unskilled jobs held by most of them, they do not earn less than future convicts, but a minority in white-collar positions are paid significantly less. Ex-prisoners' jobs are less likely to dissolve quickly if the hiring firm potentially had access to co-worker, employer, or labor office referrals.