This prospective study focused on the accuracy of diagnosis of Alzheimer's disease (AD). We recruited 100 dementia patients and 20 controls who underwent a systematic evaluation. The clinical diagnosis of probable AD or possible AD according to the NINCDS-ADRDA criteria was assigned in 69% of the patients, 21% had vascular dementia (VaD) (DSM-III-R) and 8% had mixed AD-VaD; only 2 patients (2%) had the Lewy body variant of AD (AD-LB). During a 3-year period 57 patients died, 53 of them (93%) being autopsied. Neuropathological examination according to the CERAD criteria showed definite AD in 27 out of 28 (96%) patients diagnosed as probable AD. In the possible AD group, the diagnostic accuracy was also high, 86% showed at least some degree of AD pathological alterations. The neocortical senile plaque scores correlated significantly with tangle scores in patients with AD pathology, and there was a significant negative correlation between age of onset and neocortical tangle scores. The concordance between the clinical diagnosis and pathological findings was clearly lower in VaD than in AD. In the clinical VaD group, 8 of 10 patients had at least some degree of AD changes together with vascular changes and only 2 of 10 patients had pure VaD. This study confirms the high accuracy of the NINCDS-ADRDA criteria for diagnosing AD. In contrast, uncertainty in the clinical diagnosis of VaD should be taken into account, for example, in drug trials with VaD patients.