Two experiments involving object-manipulation tasks were performed to examine whether 1-to 2-year-olds form superordinate-like categories by attending to object parts. In Study 1, 14-, 18-, and 22-month-olds were tested with contrasts of animals, furniture, insects, and vehicles. Fourteen-and 18-month-olds behaved systematically toward categories with different parts (legs or wheels) but not toward categories with matching parts (legs or legs). In Study 2, infants were tested with novel animals and vehicles generated by removing or attaching legs or wheels. In the absence of part differences, all three age groups failed to form superordinate categories. The two younger groups chose to categorize by parts (i.e., legs or wheels) rather than by category membership (animal or vehicle). The results suggest a perceptual basis for categorization whereby infants form dynamic categories, on-line, that are based on the characteristics of the input.It is a common assumption that the categories formed in early childhood fall into one of the three hierarchical classes proposed by Rosch and her colleagues (e.g., Mervis & Rosch, 1981;Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976). According to Rosch, the categories that are developmentally primary are at the basic level, with categories developing subsequently at the more general, superordinate level. This developmental sequence is thought to occur because basic-level categories have a high level of within-category similarity and between-category dissimilarity, whereas superordinate categories have low within-category similarity. Thus, it is held that infants form categories of objects that are alike, such as dogs, cats, cars, and chairs, more easily than they form categories that include objects that vary in appearance, such as animals, vehicles, and fumiture. Rosch (1978) referred to these categories as taxonomies, by which she meant that their members are the same kind of thing or are "related to one another by means of class inclusion" (p. 27).A corollary for this assumption is that infants require knowledge beyond information given in the perceptual input to form David H. Rakison and George E. Butterworth, Department of Psychology, University of Sussex, Brighton, England.This work was submitted by David H. Rakison in partial fulfilment of the requirements for the D.Phil. in psychology at the University of Sussex. We express thanks to Colin Crook and Tony Stubbens for their help in video recording and stimulus creation. We also thank Alexandra Campbell for her help with coding and Brenda Todd for her input in this research and comments on a draft of this article.The results of Experiment 1 were presented at the Meeting of the European Science Foundation, San Felui de Guixols, Spain, April 1996, and the results of Experiment 2 were presented at the Annual Meeting of the British Psychological Society Conference, Brighton, England, April 1995.Correspondence concerning this article should be addressed to David H. Rakison, who is now at the Department of Psychology, Mezes 3...