Five experiments investigated how learning about the added feature in a feature-positive discrimination or feature-negative discrimination is related to the change in reinforcement rate that the feature signals. Rats were trained in a magazine-approach paradigm with 2 concurrent discriminations between A versus AX and B versus BY. In 2 experiments (1 and 3), X and Y signaled an increase of 0.3 in the probability of reinforcement, from 0.1 to 0.4 (A vs. AX), or from 0.6 to 0.9 (B vs. BY). After extended training, each session included probe test trials in which X and Y were presented alone (Experiment 1) or in compound with another excitatory conditional stimulus (CS), C (Experiment 3). There was no difference in response rate between the 2 types of test trial (X vs. Y; XC vs. YC), consistent with the fact that X and Y signaled the same absolute change in reinforcement. In Experiments 2 and 4, X and Y signaled a decrease of 0.3 in the probability of reinforcement, from 0.4 to 0.1 (A vs. AX) or from 0.9 to 0.6 (B vs. BY). Test trials in which X or Y was presented with C showed that X had greater inhibitory strength than Y, consistent with the fact that X signaled a larger relative change in reinforcement. This was confirmed in Experiment 5, in which X and Y had the same inhibitory strength on test after training in which they signaled the same relative change in reinforcement but different absolute changes (0.3 to 0.1 for A vs. AX; 0.9 to 0.3 for B vs. BY). The results show that excitatory conditioning is linearly related to the increase in reinforcement rate, whereas inhibitory learning is not linearly related to the decrease in reinforcement rate. Implications of this for theories of associative learning are discussed.