Two attractive theories for the evolution of senescence are based on the principle that the force of natural selection decreases with age1-5. The theories differ in the type of age-specific gene action that they assume. Antagonistic pleiotropy2-5 postulates that pleiotropic genes with positive effects early in life and negative effects of comparable magnitude late in life are favoured by selection, whereas genes with the reverse pattern of action are selected against. Mutation accumulation1,3-5 assumes that deleterious mutant alleles with age-specific effects will equilibrate at a lower frequency if their effects are expressed early rather than late in life. Explicit models demonstrate that both mechanisms can lead to the evolution of senescent life histories under reasonable conditions3-5. Antagonistic pleiotropy has gained considerable empirical support4-6, but the evidence in support of mutation accumulation is more sparse4,5,7. Here we report that the genetic variability of mortality in male Drosophila melanogaster increases greatly at very late ages, as predicted by the mutation accumulation hypothesis3-5. The rate of increase in mortality with age exhibits substantial genetic and environmental variability. This result provides a possible explanation for recent observations of non-increasing mortality rates in very old flies8,9.