Recent advances in genome technology have led to an exponential increase in the ability to identify and measure variation in a large number of genes in the human genome. However, statistical and computational methods to utilize this information on hundreds, and soon thousands, of variable DNA sites to investigate genotype-phenotype relationships have not kept pace. Because genotype-phenotype relationships are combinatoric and non-additive in nature, traditional methods, such as generalized linear models, are limited in their ability to search through the high-dimensional genotype space to identify genetic subgroups that are associated with phenotypic variation. We present here a combinatorial partitioning method (CPM) that identifies partitions of higher dimensional genotype spaces that predict variation in levels of a quantitative trait. We illustrate this method by applying it to the problem of genetically predicting interindividual variation in plasma triglyceride levels, a risk factor for atherosclerosis.