Multitasking was studied in the stop-change paradigm, in which the response for a primary GO1 task had to be stopped and replaced by a response for a secondary GO2 task on some trials. In 2 experiments, the delay between the stop signal and the change signal was manipulated to determine which task goals (GO1, GO2, or STOP) were involved in performance and to determine whether the goals were activated in series or in parallel. As the delay increased, the probability of responding on stop trials changed very little, but GO2 task reaction times decreased substantially. Such effects are consistent with both a nondeterministic serial model (in which the GO1 goal is replaced by the STOP goal, which is subsequently replaced by the GO2 goal) and a limited-capacity parallel model (in which stopping and GO2 processing occur concurrently) with a capacity-sharing proportion that resembles serial processing.