One of the fundamental questions in psychology is 'How do we learn?'. Now, researchers have come a step closer to answering this question, at least for an effect called 'perceptual learning'. Practice, it appears, improves the acuity of the 'mental pictures' our brains use to represent a stimulus, such as a picture or a smell.

Perceptual learning is well known in psychology. If the brain is repeatedly required to distinguish between pictures of faces or patterns, for example, its performance improves. The same is true for other types of discrimination, such as the way wine tasters can learn to recognize the subtlest variations in flavour or bouquet.

P. J. Bennett and colleagues from the University of Toronto, Ontario, Canada, have now found evidence that repetition improves performance because it trains the brain to represent a stimulus better. This contrasts with the other prevailing theory – that training diminishes the amount of 'internal noise' or interference produced by the nervous system.

As they report in Nature1, the researchers tested these two possibilities by presenting people with pictures of faces or patterns blurred by different amounts of interference. Like the snow on a badly tuned television set, this made the images harder to see. Bennett's group then studied how practice influenced the subjects' abilities to identify the pictures.

The team found that when they added little or no artificial 'external noise' (the interference that was deliberately superimposed on the pictures), the performance of the subjects was limited only by the amount of internal noise. (Internal noise includes factors such as the degradation of an image by the optics of the eye or loss of signal quality elsewhere in the nervous system.)

As the researchers increased the amount of 'snow' on the pictures, the ability of their subjects at first remained the same. But after the interference reached a 'threshold' value, the subjects fared progressively less well as the pictures got harder to see.

Importantly, although the subjects got better at the task the more times they did it, the threshold value for external noise didn't change. This indicates that the amount of internal noise didn't change, either. Bennett and colleagues conclude that perceptual learning probably results from enhanced 'signal strength' – an improvement in the way the neurons in the brain represent a stimulus.

This conclusion has important implications for the future study of learning. It provides guidelines for the way in which computer models of the brain ought to behave, and should help researchers to search for the changes in the nervous system that occur when we learn – now that they know more about how such changes should look.