Patients with estrogen receptor (ER)-negative, basal breast cancer have a poor prognosis, and presently the best therapeutic option is chemotherapy. We have therefore modeled basal breast cancer in the mouse by combining activating mutations of β-catenin and HGF (Wnt-Met signaling). The pregnancy-induced Whey Acidic Protein (WAP) promoter was used to target Wnt-Met signaling in the mammary epithelium. Wnt-Met compound mutant mice produced rapidly growing mammary gland tumors closely resembling human basal breast cancers. We performed gene expression profiling of Wnt-Met tumors, which revealed the chemokine system CXCL12/CXCR4 as a crucial driver of Wnt-Met tumors and genetic ablation of the CXCR4 gene in Wnt-Met compound mutant mice conferred tumor resistance. We could identify an intrinsic 322 Wnt-Met mouse gene signature, which was shown to be predictive for human breast tumors. In a cohort of over 3,500 patients and 21 public datasets, the Wnt-Met signature was capable of discriminating between molecular subtypes of human breast cancer. Furthermore, the mouse Wnt-Met gene signature could predict poor survival of human patients with ER-negative breast cancers. We then applied molecular based therapy treating Wnt-Met compound mutant mice with combinations of inhibitors directed against Wnt, Met and CXCR4 signaling. Treatment combination using all three inhibitors was most effective in suppressing tumor growth. Activating both Wnt and Met signaling expands a population of cancer propagating cells. When treated with inhibitors we could show on the cellular level that the two signaling systems control different biological functions: Wnt controls self-renewal and Met controls differentiation. Taken together, our data suggests that targeting CXCR4 and its upstream activators, Wnt and Met, in cancer propagating cells, might provide a new and efficient strategy for breast cancer treatment. Furthermore, stratifying patients based on gene signatures may help accelerate clinical therapies for personalized medicine.