Insulin resistance is a heterogeneous disorder caused by a range of genetic and environmental factors, and we hypothesize that its etiology varies considerably between individuals. This heterogeneity provides significant challenges to the development of effective therapeutic regimes for long-term management of type 2 diabetes. We describe a novel strategy, using large-scale gene expression profiling, to develop a gene expression signature (GES) that reflects the overall state of insulin resistance in cells and patients. The GES was developed from 3T3-L1 adipocytes that were made "insulin resistant" by treatment with tumor necrosis factor-α (TNF-α) and then reversed with aspirin and troglitazone ("resensitized"). The GES consisted of five genes whose expression levels best discriminated between the insulin-resistant and insulin-resensitized states. We then used this GES to screen a compound library for agents that affected the GES genes in 3T3-L1 adipocytes in a way that most closely resembled the changes seen when insulin resistance was successfully reversed with aspirin and troglitazone. This screen identified both known and new insulin-sensitizing compounds including nonsteroidal anti-inflammatory agents, β-adrenergic antagonists, β-lactams, and sodium channel blockers. We tested the biological relevance of this GES in participants in the San Antonio Family Heart Study (n = 1,240) and showed that patients with the lowest GES scores were more insulin resistant (according to HOMA_IR and fasting plasma insulin levels; P < 0.001). These findings show that GES technology can be used for both the discovery of insulin-sensitizing compounds and the characterization of patients into subtypes of insulin resistance according to GES scores, opening the possibility of developing a personalized medicine approach to type 2 diabetes.