The aim of the study was to investigate the value of the Mini-Mental State Examination (MMSE) for use by general practitioners (GPs) in a group of elderly patients in whom the GPs are considering a dementia diagnosis. The study population consisted of 533 elderly patients (aged 65 and older) judged by 36 GPs as suffering from 'minimal to severe' dementia. Cross-sectional data were used to determine the criterion validity of separate items, a set of items and the total MMSE. The GMS/AGECAT diagnosis was used as an external criterion. MMSE items were analysed and two items testing general knowledge were added. The most effective set of items was determined using a stepwise logistic regression analysis. Adjusted for age, sex and education, the differentiating ability of the set of items was compared to that of the total MMSE score. The total MMSE score was divided into three categories (cutoffs 21/22 and 26/27) and into two categories (cutoff 23/24). In total, 114 patients (21%) were diagnosed as having an 'organic syndrome' by the GMS/AGECAT. The differentiating ability of separate items was poor. The following combination of items had the best predictive ability: items concerning the date, the day of the week, the patient's address and the current prime minister. This set of items was just as adequate in differentiating dementia from non-dementia as the total MMSE score (sensitivity 64.9% and 64.8% respectively, specificity 96.4% and 93.3%). The value of the MMSE in diagnosing dementia in general practice is limited. The score on cognitive test items can be one aspect of the individual's overall clinical picture, on which the diagnosis should be based.