Does semantic knowledge of objects mediate object selection and use? We present data from two patients that speak to this question. The first, DM, is a semantic dementia patient previously reported by Breedin, Saffran, and Coslett (1994) who, despite moderate to severe loss of functional and associative object knowledge, was nevertheless able to perform almost normally on single-object use and on more complex tests of naturalistic action. The second, HB, is a dementia patient who exhibited an executive disorder but performed (...) as well as controls on a detailed battery of semantic memory and single-object use tests. Unlike DM, he made numerous errors on the naturalistic action tests, among which were errors of object selection and usage. Taken together, these data suggestthatintactsemantic memory forobjects is neithernecessary norsufficient to ensure good object utilisation in naturalistic action. The data cannot be accommodated by accounts postulating that action with objects is performed exclusively via nonsemantic or visual semantic routes, but are most consistent with an account in which nonsemantic information augments deficient functional/associational semantic elements in an action-oriented network. (shrink)