Screening for dementia in populations presents particular difficulties for researchers. In the absence of gold standards for diagnosis, the methods used must be determined by the purposes of the study. In two-stage epidemiological study the screening wave and the diagnostic instrument should be considered together in relation to a third proxy gold standard such as progression of the disorder to moderate and greater severity and neuropathological diagnosis. This provides a measure of the predictive performance of the original screening method and its diagnostic phase. To reduce the variance of estimates of prevalence and incidence it is suggested that the screening interview be a subset of the diagnostic interview.