The rare early-onset variant of Alzheimer's disease (AD) appears to be transmitted as an autosomal dominant genetic trait. More typical late-onset AD also shows familial aggregation, but possible genetic mechanisms are difficult to examine because the phenotypic expression of the putative AD genotype is often censored by prior death from competing causes. Lifetable methods have been used to examine the age-specific risk of dementia among relatives, and thus to test the hypothesis of genetic transmission of late-onset AD. These methods require the ascertainment of affected relatives and the determination of their age at onset. The latter determination is somewhat arbitrary, since symptoms of AD evolve and develop in a continuous and progressive fashion, and different workers may thus use differing criteria for "onset. " This paper demonstrates that the use of divergent thresholds for caseness" (typically, progressive dementia of several years' duration) and onset (e.g., the first appearance of mild cognitive symptoms, or the first clear evidence of dementia) can introduce substantial bias toward underestimation of risk among relatives. Depending on the definition of onset, familial risk may be underestimated, with apparent cumulative incidence decreased to only 60% of values otherwise expected. We suggest that this problem can be avoided by the use of identical threshold criteria for caseness and for onset.