The procedure used with group BETWEEN also allowed us to establish a gradient of generalization of extinction. We extended the normal generalization test to an entire extinction phase by adding more nonreinforced generalization trials, so that each subgroup received extinction with one of the eight generalization test stimuli. This is expected to induce posterior discrimination training on the generalization dimension, with differing levels of complexity (from dissimilar to similar stimuli). We wondered to what extent this would have an effect on the rated outcome-expectancy to the CS itself. There have been some reports, in animal conditioning (Dubin & Levis, 1973) and in a human predictive learning task (Vervliet et al., 2006), showing a gradual recovery of outcome-expectancy to the CS after extinction with more dissimilar GSs. Such orderly recovery establishes the gradient of extinction generalization.