Backward inhibition (BI) is a performance cost that occurs when an individual returns to a task after 1 (vs. more than 1) intervening trial, and it may reflect the inhibition of task-set components during switching. In 3 experiments, we support the theory that inhibition can target cue-based preparatory stages of a task. Participants performed a cued target-localization task that had been previously shown to produce BI. In Experiment 1, reassignment of arbitrary cue-target pairings midway through the experiment doubled the size of BI, though cue, target, and response sets remained unchanged. In Experiment 2, we controlled for effects of order of conditions or simple change of cue meaning. In Experiment 3, we demonstrated that the effect depends on re-pairing members of the same cue and target sets. The results are attributed to heightened conflict (and hence greater inhibition) during cue-target translation when a previously learned cue-target mapping is remapped. (Contains 3 footnotes, 3 tables, and 3 figures.)