The effects of test stimulus range on generalization gradients in humans were assessed for discriminations between faces that varied in brightness, faces that varied in orientation in the picture plane, and morphed faces. In Experiment 1, significant range effects, predicted by adaptation level theory, occurred when faces varied along the brightness or orientation dimension, but not for morphed faces. Changing the difficulty of discrimination of both training and test stimuli for Experiment 2 produced range effects in morphed faces. Experiment 3 explored training and testing stimulus factors as determinants of range effects in morphed faces. The results suggest that sufficiently biased testing ranges create shifts in response distributions (generalization gradients), and this may be amplified by using relatively difficult discriminations between training stimuli.