There was a five-year delay between the two waves of the Canadian Study of Health and Aging during which 2,982 participants died. Their cognitive status before death should be taken into account in estimating the incidence of dementia in the cohort. Information concerning antemortem cognitive status was available from death certificates and from an interview with a close relative of the decedent at the CSHA-2 follow-up. The interview included a direct question on whether the person had been diagnosed with dementia and questions covering cognitive signs and symptoms from which we formed an algorithm to predict probability of dementia. These sources of information were validated using a small sample of study participants who died within five months of undergoing the CSHA clinical examination. Sensitivity of the death certificate and the question regarding diagnosis of dementia was low (33% and 44%), although their specificity was very high. Accordingly, we combined these with the predictive algorithm to form an overall estimate of the probability of antemortem dementia. This raised the sensitivity to 82% (specificity 93%).