The perception/action model is the dominant account of the primary division of labour in the human visual pathway. Integral to this model is the idea that goal-directed actions are guided spatially by bottom-up vision, independent of perceptual recognition and top-down object knowledge. We question this idea by showing that the expected size of familiar objects (matchboxes) affects the amplitude of reaches made to grasp them, and the pre-shaping of the hand, even when binocular cues are available. This suggests that perceptual recognition routinely influences action programming.