"Peak Shift" usually occurs following interdimensional-discrimination training and involves a shift of the peak of the generalization gradient away from the original discriminative stimulus (S+) in a direction away from an S-. Two theoretical accounts of peak shift, the gradient interaction theory (GIT) and adaptation level theory (ALT), were compared. The effects of asymmetric test stimuli and the impact of instructions to participants for them to treat stimuli as members of categories on generalization gradients were investigated. In Experiment 1, the relation between peak shifts obtained when an extended asymmetric set of test stimuli was employed and the occurrence of categorization of the stimuli involved was investigated in four separate conditions. Two involved temporal discrimination, one involved line-angle discrimination, and one involved a compound line-angle and temporal cue discrimination. If participants treated the stimuli as belonging to discrete categories, such as hands-on a clock, rather than as being on continuous dimensions then responding to the compound cue was expected to result in attenuation of blocking of a peak shift. However, the peak shift obtained to the three cue types were the same. In Experiment 2, an independent group of participants was given explicit instructions to treat the line angles as if they were the hands of a clock face and this eliminated peak shift. The results from the present experiments support an ALT interpretation, although the peak shifts were significantly smaller in magnitude than predicted by this account.