In task switching experiments participants have to respond to the same set of stimuli while task instructions vary (e.g., digit stimuli are assigned to left- or right-sided key presses by means of magnitude vs. parity classification). Response congruency effects denote worse performance for a stimulus, which is associated with different responses in the two tasks as compared to a stimulus, which is associated with the same response. Previous research suggests that such effects reflect direct links between stimuli and responses acquired in the course of experimental practice. In the current study we investigated the impact of stimulus-specific practice and task instruction by reversing the S-R mapping of one task (Experiment 1) or replacing one task with a new one (Experiment 2) in the second half of an experimental session. Consistent with the direct link account, S-R links practiced during the first half of the experiment largely determined congruency effects despite altered task instructions. Furthermore, the results suggest that previously practiced S-R links (a) can be relatively quickly overwritten by practicing a novel S-R mapping, and (b) are subject to passive decay when no longer in use.