Previous investigations suggest that there are at least three distinct types of representation of the human body. One representation codes structural information about body part location (body structural description), the second codes knowledge about body parts (body semantics or body image), and the third provides a dynamic mapping of the current positions of body parts relative to one another (body schema) (Buxbaum & Coslett, 2001; Schwoebel, Coslett, & Buxbaum, 2001; Sirigu, Grafman, Bressler, & Sunderland, 1991). In this study we used an influential "two route" model of gesture performance (Gonzalez Rothi, Ochipa, & Heilman, 1991) to derive predictions about the body representations expected to underlie the production and imitation of meaningful and meaningless movements. The relationships between these measures were examined in 55 patients with unilateral left-hemisphere lesions. Multiple regression analyses demonstrated that performance on body semantics and body schema tasks were significant and unique predictors of meaningful gesture performance, whereas the body schema measure alone predicted imitation of meaningless movements. Body structural descriptions did not enter into any of the models. These findings are consistent with performance of meaningful actions via a semantic route that accesses body semantics and other action knowledge, and performance of meaningless movements via a "direct" route that bypasses this information.