In a series of experiments, we studied the differences between natural target-directed grasping movements and 'pantomimed' movements directed towards remembered objects. Although subjects continued to scale their hand opening for object size when pantomiming, grip formation and other kinematic variables differed significantly from those seen in normal target-directed actions. This was true whether the subjects had just seen the target object 2 sec before (Experiments 1 and 2) or whether the target object was still present and they were simply required to pantomime the grasping movement beside it (Experiment 3). We argued that these pantomimed reaches were being driven by stored perceptual information about the object, and were not utilizing the normal visuomotor control systems that direct actions in real time. This interpretation received strong support from observations of a patient with visual form agnosia who was also tested. In an earlier report, we had shown that this patient showed anticipatory scaling of her grasp despite her inability to discriminate between objects perceptually on the basis of size. The present study showed, however, that the requirement to remember an object even briefly, or to pantomime an action beside it, was enough to completely disrupt her visuomotor scaling (Experiments 2 and 3). That this reflected a failure of perception rather than imagery or understanding was supported by the fact that she could convincingly pantomime actions to imagined, familiar objects, the sizes of which were known to her (Experiment 4). All these results suggest that the mechanisms underlying the formation of perceptual representations of objects are quite independent of those mediating on-line visuomotor control.