A series of four experiments measured the transfer of perceptual learning in object recognition. Subjects viewed backward-masked, gray-scale images of common objects and practiced an object naming task for multiple days. In Experiment 1, recognition thresholds decreased on average by over 20% over 5 days of training but increased reliably following the transfer to a new set of objects. This suggests that the learning was specific to the practiced objects. Experiment 2 ruled out familiarity with strategies specific to the experimental context, such as stimulus discrimination, as the source of the improvement. Experiments 3 and 4 found that learning transferred across changes in image size. Learning could not be accounted for solely by an improvement in general perceptual abilities, nor by learning of the specific experimental context. Our results indicate that a large amount of learning took place in object-specific mechanisms that are insensitive to image size.