Response congruency effects in task switching reflect worse performance for incongruent targets associated with different responses across tasks than for congruent targets associated with the same response. In the present study, the author investigated whether the effects can be produced solely by a mediated route for response selection, whereby targets are categorized with respect to both tasks, as opposed to a nonmediated route, whereby target--response instances from past experience are retrieved directly from long-term memory. The mediated route was isolated in 3 experiments by having subjects perform semantic categorization tasks on targets that were never repeated, thereby making the nonmediated route nonfunctional. Robust response congruency effects were observed for both response time and error rate in all experiments, indicating that the mediated route is sufficient to produce such effects by itself. The results imply that subjects engaged in dual-task processing despite no requirement to do so, raising questions about the modeling of response selection in task-switching situations.