The risk of hypoglycaemia and the resulting impairment in brain function are major factors preventing those with diabetes achieving normoglycaemia. The need to define these risks more precisely has prompted an increasing research effort to identify which aspects of cerebral function are particularly vulnerable and at which glucose level. Much of the evidence is inconsistent, reflecting not only the wide range of methods for measuring cognitive function and inducing hypoglycaemia, but also the inherent variability of the response. Nevertheless, the data suggest that those mental activities which are relatively undemanding are often unaffected at all levels of experimental hypoglycaemia while the performance of more complex tasks deteriorates at glucose concentrations of around 3 mmol l-1. The relative imprecision of cognitive testing is reflected in the debate which surrounds the pathogenesis of hypoglycaemia unawareness. There is some evidence that the glycaemic threshold for autonomic activation and symptoms can vary while the threshold for cognitive impairment is fixed. This has led to the hypothesis that hypoglycaemia unawareness arises when the autonomic response develops at a blood glucose below that for impaired cognition, thus preventing patients from recognizing or responding to their usual symptoms. However, contradictory data suggest that the threshold for cognitive impairment can alter in line with the autonomic response, a conclusion which falls to fit either the above hypothesis or the clinical description of hypoglycaemia unawareness. These differences may be methodological or relate to the relative imprecision of measurements of cognitive function. Resolution of these discrepancies may have to await the development of advanced technology such as high resolution MRI or PET scanning. In the meantime, progress could be made if all groups agreed on a limited range of cognitive function tests and used them in a standardized manner.