It is quite an old idea that receptive fields in the cortex are adapted, both evolutionarily and by experience, to the statistical properties of the inputs received, but it is only recently that simulations have successfully shown that this can happen. The reasons for earlier failures and later successes will be reviewed. In addition it will be argued that the main advantage derived from such an adaptive process is that it enables the cortex to predict. Prediction is a matter of pattern completion along the time dimension, and it is rather obviously a form of pattern completion that confers much greater competitive advantage than completion in spatial dimensions alone. If this is right one can make two predictions about cortical organisation: 1) spatio- temporal statistical features are much more important than pure spatial ones, and receptive fields should reflect this bias; 2) the map of the visual field in the cortex might be predictive - receptive fields may be displaced in the map along the axis they respond to optimally and by a distance that depends on their velocity tuning. Such a map would compensate for latencies in visual and motor pathways.