In 3 experiments the role of mediators in task switching with transparent and nontransparent cues was examined. Subjects switched between magnitude (greater or less than 5) and parity (odd or even) judgments of single digits. A cue-target congruency effect indicated mediator use: subjects responded faster to congruent cue-target combinations (e.g., ODD-3) than to incongruent cue-target combinations (e.g., ODD-4). Experiment 1 revealed significant congruency effects with transparent word cues (ODD, EVEN, HIGH, and LOW) and with relatively transparent letter cues (O, E, H, and L) but not with nontransparent letter cues (D, V, G, and W). Experiment 2 revealed significant congruency effects after subjects who were trained with nontransparent letter cues were informed of the relations between cues and word mediators halfway through the experiment. Experiment 3 showed that congruency effects with relatively transparent letter cues diminished over 10 sessions of practice, suggesting that subjects used mediators less as practice progressed. The results are discussed in terms of the role of mediators in interpreting instructional cues.