Inhibition in set switching is inferred from so-called n-2 repetition costs: slower response times to ABA sequences compared to CBA sequences (where A, B, and C are arbitrary labels for different tasks). These costs are thought to reflect the persisting inhibition of task A when it was disengaged recently (as is the case in an ABA sequence). In this study we were interested in whether more inhibition may be required when the tasks are relatively novel. To this end, we examined the effect of practice on the n-2 repetition cost in nine participants across five experimental sessions, with 1222 trials performed in each session. The results show a clear reduction in the n-2 repetition cost, being altogether absent from the final sessions. Such a reduction is predicted by both: (a) a recent computational model of the n-2 repetition cost (Grange, Juvina, & Houghton, 2013) due to the gradual strengthening of task-related memory elements with practice to the point where inhibition has less impact; and (b) prior work showing smaller n-2 repetition costs with greater cue-target association strength (Houghton, Pritchard, & Grange, 2009). In this paper, we integrate these two theoretical derivations by extending our computational model, which fit the current data-at the mean level, block level, and individual-subject (i.e., individual differences) level-well.