In 1992, David Milner and I (Goodale & Milner, 1992) proposed a division of labour in the visual pathways of the primate cerebral cortex between a dorsal stream specialized for the visual control of action and a ventral stream dedicated to constructing our percepts of the visual world. Support for the perception-action distinction has come from neuroimaging experiments, human neuropsychology, and monkey neurophysiology. Differences in the timing and spatial metrics of vision-for-perception and vision-for-action have been studied in human psychophysical experiments, particularly in those that have looked at the way in which each system deals with pictorial illusions. Although the literature is not free from controversy, a large number of studies have found that actions such as grasping and reaching are often unaffected by high-level pictorial illusions, which by definition affect perception. Recent experiments have shown that for actions to escape the effects of such illusions, however, they must be highly practised actions, preferably with the right hand, and must be directed in real time at visible targets. But even though the behavioural evidence suggests that the dorsal and ventral streams make use of different timing, different metrics, and different frames of reference in carrying out their computations, there is a seamless interaction between the two streams in the production of adaptive behaviour. A full understanding of the integrated nature of visually guided behaviour will require that we specify the nature of the interactions and information exchange that occur between these two streams of visual processing.