Abstract Poorly defined molecular heterogeneity and a lack of effective molecular targets hinder advancements in the treatment of ER-/PR-/HER2- breast cancer (TNBC). Meta-analyses of public datasets have revealed new subgroups, but these studies are plagued by samples collected from multiple protocols and poor reproducibility. We aimed to identify stable subsets that could be treated with targeted therapy by employing integrated genomic profiling of a large number of TNBC cases (confirmed by IHC) from a single protocol before applying those subgroups to publicly available TNBC sets. We employed non-negative matrix factorization on two independent gene expression datasets (Discovery n = 84 and Validation n = 118) using the top 1000 variant genes in each set. Cophenetic, dispersion, and silhouette metrics revealed 4 clusters, with significant enrichment of differentially expressed (DE) genes in each subgroup among both sets (all comparisons Fisher Exact p < 2.2E-16). A parsimonious gene centroid signature of 480 total genes was selected based on Goeman's Global Test BH-adjusted p-values and fold change of only the Discovery Set. Subgroups were assigned by Pearson Correlation in the Discovery (Rand Index = 0.97) and Validation Sets (RI = 0.84), as well as an External Set made up of 214 IHC-confirmed cases of TNBCs, and another 5 publicly available studies with clinical outcome data for TNBCs. Ingenuity Pathway Analysis was carried out separately on DE genes from each dataset (BH adjusted p < 0.01). There was overwhelming statistical agreement for all subgroups between the Discovery, Validation, and External Sets. Subgroup 1 was defined by Intermediate Grade LumB tumors and overexpression of ESR1, HER-2, and AR, with activated downstream signaling of ESR1 despite being ER- by IHC. In subgroup 2, CDKN2A and TP53 appeared activated while MYC was inhibited. Dozens of immune cell signaling pathways were downregulated in the third subgroup, a basal-like set of TNBCs, which was driven by inhibition of cytokine gamma-IFN. Disease-free and Overall Survival was the worst for subgroup 3 (log rank p = 0.041 and 0.039). The final subgroup (4) was found to be p53 inactivated, a known feature of TNBC but specific to this second basal-like subgroup. Analysis of 84 and 58 cases with corresponding copy number profiling data from the Discovery and Validation Sets revealed subgroup specific copy number changes in addition to known TNBC patterns. Most notably, loss of chr 6 was unique to subgroup 1, while all other tumors shared common loss of chr 5. We have described four molecular phenotypes of TNBC in two independent sets from the same protocol, and a third set composed of public data. These subgroups have strong agreement of DE genes, pathways, and regulators, and are additionally supported by DNA events and prognostic differences between subgroups. We suggest several promising druggable targets based on overexpressed genes in each subgroup. Citation Format: Matthew D. Burstein, Anna Tsimelzon, Kyle R. Covington, Suzanne AW Fuqua, Jenny C. Chang, Susan G. Hilsenbeck, C Kent Osborne, Gordon B. Mills, Ching C. Lau, Powel H. Brown. Identification of four subgroups of Triple Negative Breast Cancer (TNBC) by genomic profiling. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 1996. doi:10.1158/1538-7445.AM2013-1996