The advent of sophisticated genomic techniques for gene mapping and microarray analysis has provided opportunities to map mRNA abundance to quantitative trait loci (QTL) throughout the genome. Unfortunately, simple mapping of each individual mRNA trait on the scale of a typical microarray experiment is computationally intensive, subject to high sample variance, and therefore underpowered. However, this problem can be addressed by capitalizing on correlation among the large number of mRNA traits. We present a method to reduce the dimensionality for mapping gene expression data as quantitative traits. We used a blind method, principal components, and a sighted method, hierarchical clustering seeded by disease relevant traits, to define new traits composed of a small collection of promising mRNAs. We validated the principle of our approach by mapping the expression levels of metabolism genes in a population of F(2)-ob/ob mice derived from the BTBR and C57BL/6J strains. We found that lipogenic and gluconeogenic mRNAs, which are known targets of insulin action, were closely associated with the insulin trait. Multiple interval mapping and Bayesian interval mapping of this new trait revealed significant linkages to chromosome regions that were contained in loci associated with type 2 diabetes in this same mouse sample. As a further statistical refinement, we show that principal component analysis also effectively reduced dimensions for mapping phenotypes composed of mRNA abundances.