The effects of conditional discrimination training on human auditory frequency generalization were examined. In Experiment 1 subjects learned to press a key to one tone, but not to a higher pitched tone when presented in the left earphone, and to respond to the higher pitched tone but not to the lower tone in the right earphone. Generalization tests revealed conditional peak shifts, with subjects responding to frequencies lower than the positive stimulus (S+) for left-earphone presentations and to frequencies higher than S+ for right-earphone presentations. The shifts observed following conditional discrimination were larger than those obtained from subjects who received simple discrimination training. Experiment 2 showed that these larger shifts were not simply due to extended discrimination training. In Experiment 3 subjects were trained with conditional discriminations involving an S+ and a negative stimulus (S-) in one earphone but only a single, positive stimulus in the other earphone. These subjects also showed conditional gradient shifts with displacements in both directions. The occurrence of peak shift without an S- can only be interpreted as involving relational responding.