Most accounts of the Stroop effect (Stroop, 1935) emphasize its negative aspect, namely, that in particular situations, processing of an irrelevant stimulus dimension interferes with participants' performance of the instructed task. In contrast, this paper emphasizes the fact that, even with that interference, participants actually can (and usually do) exert enough control to perform the instructed task. An Adaptive Control of Thought–Rational (ACT–R) model of the Stroop task interprets this as a kind of learned strategic control. Specifically, the concept of (...) utility is applied to the two processes that compete in the Stroop task, and a utility‐learning mechanism serves to update the corresponding utility values according to experience and hence influence the competition. This model both accounts for various extant Stroop results and makes novel predictions about when people can reduce their susceptibility to Stroop interference. These predictions are tested in three experiments that involve a double‐response variant of the Stroop task. (shrink)