The relative incidence of the major types of dementia disorders and the agreement rates between clinical and pathological diagnosis were analysed in consecutive autopsy series of 675 demented subjects from 3 hospitals (mean age 79.5 years, SD 9.6). Clinical assessment followed the DSM-III and ICD-9-NA criteria and NINCDS/ADRDA criteria for probable Alzheimer disease (AD) (McKhann et al. 1984), histological criteria for the diagnosis of AD those of the NIH/AARP Work Group (Khachaturian 1985) using a 4-degree rating scale for plaques and tangles in neocortex and hippocampus (Morris et al. 1988), and the criteria by Tierney et al. (1988) for 'pure' AD. Vascular dementia (MID) and other disorders were diagnosed according to current pathologic criteria. Clinical diagnosis of AD/SDAT was made in 59.2%, of MID in 21.7%, of mixed AD + MID in 3.1%, and of Parkinson's disease (PD) and other disorders in 16%. At autopsy, 76.7% fulfilled histological criteria for AD/SDAT, but only 60% were 'pure' forms, while 8.2% had additional features of PD and 7.9% coexisting vascular lesions indicating mixed SDAT + MID. 15.7% were MID with no or very little AD pathology, 7.4% other CNS disorders. 0.3% of the brains showed no abnormality beyond age-related changes. AD/SDAT had its highest incidence in a psychiatric population, MID and PD + SDAT in general and geriatric hospital cohorts. The overall coincidence rates for clinical and pathological diagnosis of AD/SDAT were 85.2%, for MIX and MID 60.5-61.9%, but only 51% for PD-PD/AD. These data and the results of other recent studies emphasize the need for more appropriate clinical and pathological criteria in the diagnosis of the dementias.