In Experiment 1, rats were trained on a discrimination in which one occasion setter, A, signaled that one cue (conditioned stimulus, CS), x, would be followed by one outcome, p (unconditioned stimulus, US), and a second CS, y, by a different outcome, q (x [right arrow] p and y [right arrow] q); a second occasion setter, B signalled the reverse CS-outcome relations (x [right arrow] q and y [right arrow] p). In a subsequent stage, the animals were divided into two groups, and trained as before, except that both A and B were presented in compound with a novel occasion setter, C. For Group S (same) the CS-outcome relations following A and B were identical to those in the pretraining stage, whereas in Group D (different) they were reversed. In a subsequent test, stimulus C was shown to be a more effective occasion setter in Group D than in Group S. In Experiment 2, rats were trained on a negative occasion-setting discrimination in which CS x signaled outcome p, and y outcome q; when x and y were signaled by the occasion setter A then no outcome followed (x [right arrow] p, y [right arrow] q, A:x-, and A:y-). In a subsequent stage, A was now trained as a positive occasion setter, signaling reinforcement of x and y. In Group S, x and y signaled the same outcomes as in the prior training stage (x-, y-, A:x [right arrow] p, and A:y [right arrow] q), whereas in Group D they signaled the opposite outcomes (x, y, A:x [right arrow] q, and A:y [right arrow] p); more efficient test performance was seen in the latter group. These results suggest that the each occasion setter conveyed information about the specific combination of CS and US paired in its presence (i.e., x [right arrow] p and y [right arrow] q, or x [right arrow] "no" p and y [right arrow] "no" q). These results are consistent with the suggestion that occasion setters operate, at least in part, on a specific CS-US association.