The taxonomic assumption, or noun-category bias, is thought to facilitate word learning by focusing children's attention on taxonomic categories as likely candidates for word meanings. Three experiments were conducted to disentangle the role of taxonomic relations and shape similarity in 18- and 24-month-olds' responses on a noun category bias task. The relationship between vocabulary composition and performance on these tasks was also examined. Results indicated that both 18- and 24-month-old children were predominantly guided by shape similarity when extending novel labels. However, some evidence that taxonomic information can be used to guide word extension was found in the 24-month-old group. Those children with a larger proportion of nouns in their vocabulary were more likely to use information about category membership when extending words, even in the absence of shape similarity.