Background: Early detection of breast cancer provides the best opportunity for cure. Mammography is the benchmark for screening, but suffers technical, logistic, and diagnostic limitations. An effective and accurate blood test to detect early stages of the disease should increase the screening detection rate for breast cancer. We conducted a series of lipidomic studies in early-stage breast cancer patients and combined the datasets via a machine learning driven analysis to test if plasma lipidomic profiles can detect breast cancer. Methods: Blood samples were collected from women with stage 0-IV breast cancer (4 separate cohorts) with age and BMI matched breast cancer-free controls. Lipids from plasma enriched extracellular vesicles were extracted and analysed by high resolution accurate mass LC-MS. A commercially available software was used to annotate and quantify >400 manually curated lipid species. Following variable selection, a lipid signature was identified capable of distinguishing breast cancer samples from control. Results: Plasma samples from women with breast cancer were distinguished from controls with an average cross-validated accuracy of 0.81, and average AUC of 0.84 across 4 cohorts (Table 1). An optimised cross-cohort subset of early-stage IDC, DCIS and ILC were differentiated from controls with a cross-validated AUC of 0.90, sensitivity of 0.88 and specificity of 0.82 (201 early-stage breast cancer, 199 controls) (Table 1). For this optimised cohort our test achieved a sensitivity of 0.71 at a prescribed specificity of 0.90, or equivalently a sensitivity of 0.89 at a prescribed specificity of 0.80. Conclusions: Our study demonstrates the high sensitivity and specificity of a lipid biomarker signature with potential for early detection of breast cancer. Ongoing studies will prospectively compare the lipid-biomarker based test against mammographic and pathological diagnosis.