reached the criterion of 90% correct in two successive sessions; he showed no evidence of transfer whatever (to an inverted triangle and a vertical line). Clea was trained with a sample set of white dot and a circle and was tested for generalization with the red disk and inverted triangle. Atom received initial training with the red disk and an upright triangle and was tested with the dot and the circle. Although Clea showed no evidence of transfer on the first 24 test trials with the new sample set, during the subsequent three test sessions her performance on the test set was indistinguishable from her performance on the old stimuli. Atom showed significant (p 〉 .01) generalization to the new test pair during the very first 24 trials, his performance increasing thereafter. Such rapid generalization of identity matching is impressive in view of the fact that original training required hundreds of trials and taking into account the fact that, because the animals had not previously seen the test stimuli, some degree of stimulus learning had to occur before they could match successfully. It should also be pointed out that, unlike Atom, who was trained and tested with simultaneous matching, Clea had been trained with and tested with 0-sec delay matching. Consequently, her poor performance during the first 24 trials might be due in part to a memory decrement associated with the new sample set.
A reasonably comparable assessment of generalization of identity matching was made with pigeon subjects. The birds learned the initial