We conducted three experiments to investigate the associative structure underlying the reinstatement of instrumental performance after extinction. In each experiment, rats were initially rewarded on two responses with different outcomes. At test, both responses were extinguished in order to assess the impact of a single noncontingent outcome delivery on response selection. Experiment 1 found evidence of outcome-selective reinstatement (i.e., more responses were performed on the lever that was trained with the reinstating outcome than on the other lever). Experiment 2 demonstrated that the outcome's capacity to reinstate performance was not affected by a reduction in its motivational value. Experiment 3 found evidence that the reinstating outcome selectively retrieved the response it signaled rather than the response it followed during training. Together, these findings are consistent with the view that instrumental reinstatement depends on the discriminative stimulus properties of the reinstating outcome.