Individuals with autism spectrum disorders (ASDs) have been assumed to show evidence of abnormal visuospatial processing, which has been attributed to a failure to integrate local features into coherent global Gestalts and/or to a bias towards local processing. As the available data are based on baseline performance only, which does not provide insight into cognitive/neural plasticity and actual cognitive potential, we investigated how training-resistant possible visuospatial processing differences between children with and without ASD are. In particular, we studied the effect of computerized versus face-to-face visuospatial training in a group of normally intelligent children with ASD and typically developing children as control. Findings show that (a) children with and without ASD do not differ much in visuospatial processing (as assessed by a tangram-like task) and the few differences we observed were all eliminated by training; (b) training can improve visuospatial processing (equally) in both children with ASD and normally developing children; and (c) computer-based and face-to-face training was equally effective.