After training under short or long fixed‐interval schedules, humans responded under a modified fixed‐interval schedule in which magnitude of reinforcement (X or 2X) was minimally correlated with response frequency. Response frequencies that equaled or exceeded a minimum response criterion were followed by the larger reinforcer at the end of the interval; otherwise, the smaller reinforcer was delivered. The modified schedule alternated with the baseline schedule across conditions. In a control condition, the reinforcer magnitudes produced by control subjects were yoked to those of experimental subjects. Experimental subjects, but not control subjects, showed increased responding. In addition to the baseline and modified fixed‐interval schedules used in Experiment 1, subjects in Experiment 2 also responded under a second modified fixed‐interval contingency in which increases in reinforcer magnitude were more highly correlated with response frequency. Experimental subjects, but not control subjects, showed increased responding under both procedures. Direct comparison of these two procedures showed that the high‐correlation procedure produced greater increases in responding than did the low‐correlatio