Deficits in reward decision-making are thought to contribute to criminal offending. These impairments have been measured in laboratory studies using the Iowa Gambling Task (IGT) which assesses implicit learning of different reward/punishment contingencies. This study compared IGT performance between a sample of justice-involved individuals and community-based individuals without an offending history. Participants included 100 adults from two Canadian federal correctional institutions (34% female, Mage = 39.14 ± 9.74) and a comparison group of 89 community adults with no history of offending (39% female, Mage = 37.04 ± 10.79). Responses on the IGT were analyzed for overall net score, learning across the task, and deck switching patterns. Associations between IGT performance and sentence characteristics and static factors assessment of recidivism risk were examined for the justice-involved group. The justice-involved group performed significantly worse than community adults in terms of net score. While the community group learned the advantageous strategy across the task, justice-involved participants exhibited minimal learning. This effect was moderated by recidivism risk within the justice-involved group, with individuals at low risk, but not medium/high risk, showing improvement over the blocks of the task. Finally, the justice-involved group also made greater use of an ineffective “win-stay/lose-shift” strategy. These results suggest that, compared with community participants without history of offending, incarcerated adults tend to employ maladaptive decision-making strategies that yield worse overall outcomes and the extent of impairment is associated with recidivism risk.