One of the most controversial issues in the cognitive neuroscience literature on concepts is whether the motor features of verb meanings are represented in the precentral motor cortices. Much of this debate stems from the fact that the empirical data are mixed with regard to (1) whether action verbs engage the precentral motor cortices in the predicted ways, (2) whether that engagement is automatic, and (3) whether it is essential for comprehension. I argue that the available data can best be accommodated by theoretical models which assume that conceptual knowledge is underpinned by a flexible, multilevel architecture that includes not only low-level modality-specific systems for perception, action, and emotion, but also high-level cross-modal convergence/divergence zones, as well as the statistical cooccurrence patterns of word-forms across discourses. From the perspective of such pluralistic approaches, the motor features of verb meanings are indeed represented in the precentral motor cortices, but their retrieval is modulated by task and context and is not always necessary for word comprehension.