Detecting gene-gene interaction in complex diseases is a major challenge for common disease genetics. Most interaction detection approaches use disease-marker associations and such methods have low power and unknown reliability in real data. We developed and tested a powerful linkage-analysis-based gene-gene interaction detection strategy based on conditioning the family data on a known disease-causing allele or disease-associated marker allele. We computer-generated multipoint linkage data for a disease caused by two epistatically interacting loci (A and B). We examined several two-locus epistatic inheritance models: dominant-dominant, dominant-recessive, recessive-dominant, recessive-recessive. At one of the loci (A), there was a known disease-related allele. We stratified the family data on the presence of this allele, eliminating family members who were without it. This elimination step has the effect of raising the "penetrance" at the second locus (B). We then calculated the lod score at the second locus (B) and compared the pre- and post-stratification lod scores at B. A positive difference indicated interaction. We also examined if it was possible to detect interaction with locus B based on a disease-marker association (instead of an identified disease allele) at locus A. We also tested whether the presence of genetic heterogeneity would generate false positive evidence of interaction. The power to detect interaction for a known disease allele was 60-90%. The probability of false positives, based on heterogeneity, was low. Decreasing linkage disequilibrium between the disease and marker at locus A decreased the likelihood of detecting interaction. The allele frequency of the associated marker made little difference to the power.