Insulin resistance (IR) affects a quarter of the world’s adult population and is a major factor in the pathogenesis of cardio-metabolic disease. In this pilot study, we implemented a non-invasive breathomics approach, combined with random forest machine learning, to investigate metabolic markers from obese pre-diabetic Hispanic adolescents as indicators of abnormal metabolic regulation. Using the ReCIVA breathalyzer device for breath collection, we have identified a signature of 10 breath metabolites (breath-IR model), which correlates with Homeostatic Model Assessment for Insulin Resistance (HOMA-IR) (R = 0.95, p < 0.001). A strong correlation was also observed between the breath-IR model and the blood glycemic profile (fasting insulin R = 0.91, p < 0.001 and fasting glucose R = 0.80, p < 0.001). Among tentatively identified metabolites, limonene, undecane, and 2,7-dimethyl-undecane, significantly cluster individuals based on HOMA-IR (p = 0.003, p = 0.002, and p < 0.001, respectively). Our breath-IR model differentiates between adolescents with and without IR with an AUC-ROC curve of 0.87, after cross-validation. Identification of a breath signature indicative of IR shows utility of exhaled breath metabolomics for assessing systemic metabolic dysregulation. A simple and non-invasive breath-based test has potential as a diagnostic tool for monitoring IR progression, allowing for earlier detection of IR and implementation of early interventions to prevent onset of type 2 diabetes mellitus.