New methods are needed for the identification of pathogenic alleles of candidate genes that may increase cancer susceptibility. Such risk alleles are expected to be of low penetrance and may act alone or modify the effects of other genes. We have developed a method that enriches for pathogenetic disease variants contingent on gene–gene interactions. Candidate gene pairs are chosen based on previous evidence demonstrating genetic or biochemical interaction. Utilizing a cohort of sibling pairs affected with breast cancer, we tested this paradigm by identifying risk variants of cell cycle control gene CDKN1A by means of interactions with TP53 and BRCA1. This approach, which we call disease association by locus stratification, first stratified affected pairs based on sharing of both alleles at a microsatellite marker linked to CDKN1A. The second stratification was based on microsatellite marker sharing at TP53 or BRCA1. We identified subsets of affected pairs sharing both alleles at both loci as screening targets. Utilizing this approach, we were able to enrich for two noncoding disease haplotypes of CDKN1A by virtue of BRCA1 interactions. We defined each haplotype by single-nucleotide polymorphisms at two positions potentially important in CDKN1A transcriptional activation by both p53 and BRCA1p. Our results indicated that an approach based on allele sharing and gene–gene interactions will be valuable not only in identifying risk alleles but also in elucidating their mechanism of action.