The epsilon 4 allele of the apolipoprotein E locus (APOE) has been found to be an important predictor of Alzheimer disease (AD). However, linkage analysis has not clarified the role of APOE in the transmission of AD. The results of the current study provide evidence that the pattern of transmission of memory disorders differs in nuclear families in which the AD-affected proband did carry an epsilon 4 allele versus those families in which the AD-affected proband did not carry an epsilon allele. Further, risk of AD due to APOE genotype in the probands is modified by family history of memory disorders, suggesting gene-by-gene interactions. Family history remained a significant predictor of AD for affected probands with some, but not all, APOE genotypes in a logistic regression analysis. Though nonadditive in the prediction of AD, APOE genotype and family history acted additively in the prediction of age at AD onset. The results of complex segregation analysis were inconsistent with Mendelian segregation of memory disorders both in families of affected probands who did or did not carry an epsilon 4 allele, yet these two groups had significantly different parameter estimates for their transmission models. These results are consistent with gene-by-gene interactions, but also could result from common elements in the familial environment.