Systematic variations of the block design task were given to 20 autistic, 33 normal and 12 mildly retarded subjects. Designs were contrasted which were either "whole" or segmented, rotated or unrotated, and which did or did not contain obliques. Only segmentation, but neither of the spatial orientation factors, revealed a significant group difference. Autistic subjects, regardless of age and ability, performed better than controls when presented with unsegmented designs. This result suggests that they need less of the normally required effort to segment a gestalt, and thus supports the hypothesis of weak central coherence as a characteristic of information processing in autism.