Article abstract-Objective: To evaluate the predictive validity and temporal stability of diagnostic criteria for mild cognitive impairment (MCI). Background: MCI has been proposed as a nosologic entity referring to elderly persons with subclinical cognitive deficits due to incipient dementia. Classification criteria, which have been derived from small, selected clinical groups, are currently disputed, and have not yet been assessed within the general population. Methods: Subjects meeting current criteria for MCI and also age-associated cognitive decline (AACD-a similar concept that is assumed to be related to normal cognitive aging processes rather than incipient dementia) were identified within each of three waves of a longitudinal population study, which included a standardized neurologic examination. Results: In the general population, the prevalence of MCI was estimated to be 3.2% and AACD 19.3%. MCI was a poor predictor of dementia within a 3-year period, with an 11.1% conversion rate. Subjects with MCI also constituted an unstable group, with almost all subjects changing category each year. Discriminant function analysis failed to isolate a homogeneous clinical group. Subjects classified as AACD, contrary to the theoretical assumptions underlying the disorder, represented a more stable group, with a 28.6% conversion rate to dementia over 3 years (relative risk ϭ 21.2). Conclusion: MCI criteria perform poorly when applied to a representative population sample. The authors propose modifications to current diagnostic criteria to increase their capacity to detect incipient dementia.