Specifically-language-impaired children and younger normal children matched for expressive language were presented with unfamiliar object names and referents across five experimental sessions. The objects differed in the degree to which they were associated with actions, and only certain of the object exemplars were named during presentation. Comprehension testing revealed that the specifically-language-impaired children acquired a greater number of object concepts presented in a no-action condition than the normal children. However, their extension of the names to new exemplars was more restricted and less differentiated. Several possible accounts of these findings are evaluated.