The traditional approach in case-control association studies, to evaluate candidate genes individually, either single markers or haplotypes, has had limited success. The multifactorial nature of complex diseases suggests the alternative of examining gene-gene interactions (epistasis) within biological networks. We have used synergy factor analysis to assess over 100 claims of epistasis in sporadic Alzheimer's disease (AD), in networks involving, e.g. cholesterol, beta-amyloid, inflammation and oxidative stress. We found 27 gene-gene interactions that were significantly associated with AD. In most of these the main effect of one of the genes was so small that it would probably have been missed by the traditional locus-by-locus approach. There are questions, however, about the quality of replication studies: about sample sizes, about homogeneity, characterization and matching of sample sets, and about the statistical methods commonly used to assess interactions. Meta-analyses could now be conducted of the four interactions that have been sufficiently replicated, all involving APOE4: ACT -17AA; BACE1 exon5 GG; IL6 -174C; BCHE K. Only that between BACE1 exon5 GG and APOE4 has so far been consistently replicated. We conclude that epistasis is a crucial feature of complex diseases, that its study is a promising approach to the genetics of AD and that larger and more rigorous studies are needed to establish interactions.