The practice of using discrete clinical diagnoses in genetic association studies has seldom led to a replicable genetic model. If, as the literature suggests, weak genotype-phenotype relationships are detected when clinical diagnoses are used, power might be increased by exploring more fundamental biological traits. Emerging solutions to this include directly modeling levels of the protein product of a gene (usually in plasma) and sequence variation specifically in/around that gene, as well as exploring multiple quantitative traits related to a disease of interest. Here, we attempt a strategy based upon these premises examining sequence variants near the TNF locus, a region widely studied in cardiovascular disease. Multilocus genotype models were used to perform a systematic screen of 18 metabolic and anthropometric traits for genetic association. While there was no evidence for an effect of TNF polymorphism on plasma TNF levels, a relatively strong effect on plasma PAI-1 levels did emerge (P=0.000019), but this was only evident in post-myocardial infarction patients. Modeled jointly with the common 4G/5G insertion/deletion polymorphism of SERPINE1 (formerly PAI), this effect appears large (10% of variance explained versus 2% for SERPINE1 4G/5G). We exhibit this finding cautiously, and use it to illustrate how transitioning the study of disease risk to quantitative traits might empower the identification of functionally variable genes. Further, a case is highlighted where association between sequence variation in a gene and its product is not readily apparent even in large samples, but where association with a down-stream pathway may be.