The accuracy of a dementia diagnosis by specialist physicians, as verified at an autopsy, is greater than 90% in many series. Donations of brains to the Maritime Brain Tissue Bank (MBTB) by individuals who did not have expert dementia diagnoses before death led us to investigate whether clinical features could also be detected retrospectively. Informants for 36 individuals whose brains were in the MBTB (18 women, mean age = 79 years; pathologic diagnoses: 75% Alzheimer's disease [AD]; 8.4% vascular or mixed dementia) were interviewed by specialist physicians using a semistructured retrospective interview based on the Brief Cognitive Rating Scale (BCRS) (range = 1 [no impairment] to 7 [terminal dementia]). The mean duration of dementia was 8.5 ± 12.8 years based on proxy reports, and most cases suggested severe dementia—(stage 6 [severe] or 7 [terminal])—on the retrospective BCRS (RetroBCRS) before death. A score of 4 or more on the RetroBCRS had 100% sensitivity and specificity in detecting dementia. The RetroBCRS score correlated moderately with duration (.51). In linear and logistic regression models adjusted for age and sex, RetroBCRS staging helped explain 93% of the variation in duration. The accuracy of the retrospective diagnosis of the cause of dementia, compared with autopsy, was 92%. The RetroBCRS used by an expert physician with a reliable informant is a valid method of detecting dementia and determining whether AD was present.