Neuropsychological studies indicate that, after brain damage, the ability to imitate meaningful or meaningless actions can be selectively impaired. However, the neural bases supporting the imitation of these two types of action are still poorly understood. Using PET, we investigated in 10 healthy individuals the neural mechanisms of imitating novel, meaningless actions and familiar, meaningful actions. Data were analyzed using SPM99. During imitation, a significant positive correlation (p < .05, corrected) of regional cerebral blood flow with the amount of meaningful actions was observed in the left inferior temporal gyrus only. In contrast, a significant positive correlation (p < .05, corrected) with the amount of meaningless movements was observed in the right parieto-occipital junction. The direct categorical comparison of imitating meaningful (100%) relative to meaningless (100%) actions showed differential increases in neural activity (p < .001, uncorrected) in the left inferior temporal gyrus, the left parahippocampal gyrus, and the left angular gyrus. The reverse categorical comparison of imitating meaningless (100%) relative to meaningful (100%) actions revealed differential increases in neural activity (p < .001, uncorrected) in the superior parietal cortex bilaterally, in the right parieto-occipital junction, in the right occipital-temporal junction (MT, V5), and in the left superior temporal gyrus. Increased neural activity common to imitation of meaningless and meaningful actions compared to action observation was observed in a network of areas known to be involved in imitation of actions including the primary sensorimotor cortex, the supplementary motor area, and the ventral premotor cortex. These results are compatible with the two-route model of action imitation which suggests that there are at least two mechanisms involved in imitation of actions: a direct mechanism transforming a novel action into a motor output, and a semantic mechanism, on the basis of stored memories, that allows reproductions of known actions. Our results indicate that, in addition to shared neural processes, the direct and the semantic mechanisms that underlie action imitation also draw upon differential neural mechanisms. The direct mechanism underlying imitation of meaningless actions differentially involves visuospatial transformation processes as evidenced by activation of areas belonging to the dorsal stream. In contrast, imitation of meaningful actions differentially involves semantic processing as evidenced by activation of areas belonging to the ventral stream.