In bottleneck models of overlapping-task performance, stimulus–response translation for secondary tasks is postponed until the primary response is selected. If this is so, then compatibility between the secondary and primary responses, or between the secondary response and the primary stimulus, should not affect primary-task performance. Yet such effects were demonstrated in 5 dual-task experiments combining primary manual and secondary vocal tasks: Pronounced effects of compatibility between the secondary and primary response and between the secondary response and primary stimulus were found on primary-task performance. The latter effect was also found with the lowest level of an extensive stimulus onset asynchrony variation, when the secondary task was not speeded, and even when the 2 tasks were performed on different trials. Findings suggest that secondary responses were activated before primary response selection was completed and thus support an automatic-translation hypothesis holding that, although eventual response selection may be serial, stimulus–response translation is performed in parallel.