The Mini-Mental State Examination was administered to 1865 general-practice patients aged 75 years and over. Even when demented cases were removed from analysis, respondents with relatively little education, together with those in social classes III-manual and below, were significantly more likely to score below the cut-off point used in North American community surveys to denote 'cognitive impairment'. Education and social class influenced scores on all sections within the MMSE with the exception of registration. Sex influenced scores on tests of calculation and spelling backwards but had no effect on total scores. These findings emphasize the importance of investigating low scorers in more detail before making a diagnosis of dementia.