In this study we analysed the accuracy of two sets of clinical diagnostic criteria, the NINCDS/ADRDA and DSM-III-R, in relation to the currently used pathological diagnostic criteria for Alzheimer's disease (AD), the Khachaturian criteria, the Tierney A3 criteria and the CERAD protocol. The sensitivity of the individual clinical diagnostic criteria, NINCDS/ADRDA and DSM-III-R, is poor (34-58%) irrespective of the pathological diagnostic criteria applied for the definite diagnosis of AD. The combination of the NINCDS/ ADRDA 'possible' and 'probable dementia of the Alzheimer type' (DAT) categories has a high sensitivity (91-98%). However the combination resulted in very poor specificity (40-61 %). Thus, none of the clinical diagnostic criteria is satisfactory. We found similar results when we analysed the predictive value of these clinical diagnostic criteria. The positive predictive value of NINCDS 'probable DAT' category and that of the DAT diagnosis by DSM-III-R is very high (89-100%). This makes the use of these categories suitable for research purposes. However, the negative predictive value of both diagnoses is poor (33-63%), making these criteria unsuitable for diagnostic purposes in clinical practice.