Replication of statistically significant associations between single nucleotide polymorphisms (SNPs) and disease phenotypes has been problematic. One reason for conflicting observations may be failure to consider confounding factors, including gene-gene (epistatic) interactions. Our experience with the insertion/deletion polymorphism at -688 in the promoter region of plasminogen activator inhibitor (PAI-1) seems to support this contention and may foreshadow problems for genome-wide association scans, which tend to use unadjusted analytical methodologies. One hundred forty-nine patients with ≥15% total body surface area (TBSA) burns, without significant nonburn-related trauma (injury severity score ≤16), traumatic or anoxic brain injury or spinal cord injury who survived >48 hours postadmission were enrolled under a protocol approved by the UT Southwestern and Parkland Hospital IRBs. Clinical data were collected prospectively and candidate polymorphisms in PAI-1 (-688), toll-like receptor 4 (+896), CD14 (-159), tumor necrosis factor-α (-308), and interleukin-6 (-174) were genotyped. The PAI-1 SNP was significantly associated (P-value for trend = 0.036) with risk for death when evaluated in isolation by unadjusted analysis. However, after adjustment for potential confounders using multiple logistic regression, only age, full-thickness burn size, and CD14 genotype (as previously reported) were associated with increased mortality. Genetic association analyses should be adjusted for interactions between multiple SNPs, injury or disease characteristics, and demographic variables. Increasingly sophisticated analytical methods will be required as gene-mapping studies transition from a candidate-gene based approach to genome-wide association scans.