During the last decade much attention has been paid to methodological differences as an important source of variability in the results of prevalence studies of dementia.1-3 Application of internationally accepted criteria for dementia and standardized diagnostic procedures have indeed been effective in diminishing the variation in prevalence rates of studies conducted recently. Based on meta-analysis of such studies the prevalence of dementia among the Dutch population age 65 and over can be estimated to range from 5.1% to 6.7%. [4][5][6] However, the issue of non-response, a potentially important source of variation, has attracted relatively little attention. Most studies show that non-responders do not differ from responders in terms of sociodemographic characteristics. However, a similar distribution of demographic variables in responders and nonresponders does not preclude the possibility of an unequal distribution of cognitive status, 2,7-9 which is not unlikely to affect response. 10,11 Under-or overrepresentation of cognitively impaired subjects among responders may have important implications for prevalence estimates. This is especially true in studies with low response rates, but it can also be a serious problem in studies with high response rates and large differences in cognitive status between responders and non-responders. Moreover, non-responders do not necessarily constitute a homogeneous group.12 With regard to cognitive functioning, it is unlikely that refusers, those not responding due to health problems or those moving out of the study region, are comparable. Next to response rates it can thus be hypothesized that the impact of non-response on prevalence estimates of dementia depends on the type of non-response. Background. Differential distributions of sociodemographic characteristics and cognitive impairment in responders and non-responders may result in a biased prevalence estimate of dementia based on responders only. Methods. Responders (n = 2191) to a cross-sectional, two-stage community study were compared with regard to sociodemographic characteristics and cognition with three subgroups of non-responders: (A) subjects who refused to participate (n = 369), (B) subjects who were too ill or who had died prior to the screening (n = 72) and (C) subjects who had moved out of the study region or were not traceable (n = 23). Prevalence estimates specific for age and housing situation in responders and physicians' ratings of cognitive impairment were used to estimate the prevalence of dementia among non-responders.Results. Group A differed from responders in age and housing situation, group B in age, housing and cognition, and group C only in age. Separate prevalence estimates of dementia based on age, housing and cognition yielded figures for group A between 4.9% and 7.2%, for group B between 13.1% and 19.1%, and for group C between 2.6% and 4.2%. Joined with the prevalence rate among responders (6.5%) the best possible point estimate of the prevalence of dementia in the target population lies...