In this article I discuss the thesis according to which understanding language requires, at least in some cases, mental simulation. The plausibility of this «simulative theory of understanding» is assessed by a twofold line of investigation. In the first part I discuss whether the theory is actually vindicated by the available empirical evidence. In the second part, I propose a conceptual analysis of the notions of simulation and imagery. The two branches of my investigation will take us to one and the same conclusion: the simulative theory can be defended only in a weak version according to which understanding "certain classes" of sentences requires the activation of sensorimotor representations. In other words, sensorimotor representations are just an "aspect" of our conceptual abilities.