The contingency-shifting variant Iowa Gambling Task (IGT), in which the reward and punishment contingencies of different decks of cards are systematically altered, was investigated with a large group of healthy young adults (n = 208). Our findings demonstrate that the onset of unsignaled, contingency-shift phases initially disrupted learning but that performance subsequently improved during each shift. Subjective experience ratings were positively correlated with performance across all phases. A regression model showed that performance early in the task, in Blocks 3 and 4, significantly predicted later ability to shift to the changing contingencies. Subdividing participants into high performer and low performer groups revealed an increased number of selections of previously good-now-bad decks in the latter group. Overall, the contingency-shifting variant IGT may have potential as a novel measure of reversal learning in experimental and clinical settings.