* Objective * Although tool use disorders are frequent in neurodegenerative diseases, the question of which cognitive mechanisms are at stake is still under debate. Memory-based hypotheses (i.e., the semantic knowledge hypothesis and the manipulation knowledge hypothesis) posit that tool use relies solely on stored information about either tools or gestures whereas a reasoning-based hypothesis (i.e., the technical-semantic hypothesis) suggests that loss of semantic knowledge can be partially compensated by technical reasoning about the physical properties of tools and objects. * Method * These three hypotheses were tested by comparing performance of 30 healthy controls, 30 patients with Alzheimer's disease and 13 patients with semantic dementia in gesture production tasks (i.e., pantomime of tool use, single tool use, real tool use) and tool or gesture recognition tasks (i.e., functional and contextual matching, recognition of tool manipulation). Individual, item-based patterns of performance were analyzed to answer the following question: Could participants demonstrate the use of tools about which they had lost knowledge? With this aim in mind, “validation” and “rebuttal” frequencies were calculated based on each prediction. * Results * Predictions from the technical-semantic hypothesis were more frequently observed than memory-based predictions. A number of patients were able to use and demonstrate the use of tools for which they had lost either semantic or manipulation knowledge (or both). * Conclusions * These data lead to question the role of different types of memory in tool use. The hypothesis of stored, tool-specific knowledge does not predict accurately clinical performances at the individual level. This may lead to explore the influence of either additional memory systems (e.g., personal/impersonal memory) or other modes of reasoning (e.g., theory of mind) on tool use skills.