To provide a differentiated characterization of autistic children's logical deficits, a non-verbal, microanalytic method designed for infants and young children was used. Subjects were 12 autistic children ranging in age from 5 to 7 years and a control group of 12 normal children ranging in age from 23 to 30 months. Subjects were given two classification tasks: a one-dimensional and a two-dimensional task. Each task began with a free-play condition followed by provoked sorting and modeling conditions. Autistic and control children were matched on the basis of their ability to sort classes based on similarity in a free-play condition of the one-dimensional task. Comparisons were then made on other aspects of their free-play activity and on their performance in the provoked and modeling conditions in both tasks. This paper reports results of a comparison of subjects' activities in the provoked sorting conditions. Autistic children were found to be greatly delayed in their logical development. They were unable to coordinate the construction of two classes simultaneously. Autistic children engaged in fewer relevant sorting placements than normal children. In contrast, however, when only the relevant sorting placements were considered, autistic children produced a higher percentage of correct classification placements than controls. Results suggest that the autistic children accommodated more to the classification structured of the objects to be sorted than did the normal children in the conditions containing objects with multiple perceptual attributes. Implications of the findings are discussed. (RH)