The role of memory in the Iowa Gambling Task (IGT) was tested in two experiments that dissociated item memory (memory for losses obtained) from source memory (the deck that produced a given loss). In Experiment 1, participants observed 75 choices that had been made by controls or patients in previous research, followed by memory tests, and then 25 active choices from the participant. In Experiment 2, participants made choices for 75 trials, performed the memory tests, and then made 25 final choices. The data show that item and source memory can diverge within the IGT, and that source memory makes a significant contribution to IGT performance.