We use rich microdata on bank robberies to estimate individual-level disutilities of imprisonment. The identification rests on the money versus apprehension trade-off that robbers face inside the bank when deciding whether to leave or collect money for an additional minute. The distribution of the disutility of prison is not degenerate, generating heterogeneity in behavior. Our results show that unobserved heterogeneity in ability is important for explaining outcomes in terms of haul and arrest. Furthermore, higher ability robbers are found to have larger disutilities, suggesting that increased sentence lengths might effectively target these more harmful criminals.