Over 3 years we followed 8 pairs of male twins one or both of whom had suspected Alzheimer's disease (AD) including 'mild/ambiguous' changes suggestive of incident AD. These pairs were screened in 1988 and 1989 from 339 pairs in the (US) National Academy of Sciences-National Research Council Registry (NASR) of aging veteran twins, then 61-72 years of age. Most of the suspected cases (10 of 12) had mild/ambiguous changes. Including these subjects, we had estimated the prevalence of AD in the NASR as about 2%. We now describe briefly the longitudinal evaluation of these 8 pairs. Only 1 of the 10 individuals with mild/ambiguous changes has progressed to show well-defined clinical symptoms of AD. Two others remain in their original research category, while 7 clearly do not have AD. Thus, we now estimate the 1988-1989 prevalence of AD in the NASR as 0.5%. These results contrast with other follow-up studies of mild cases from a university-based Alzheimer's clinic. We suggest that the contrasting findings reflect the nature of the samples studied, and we show that the present results are predicted by Bayesian reasoning.