Malignant breast tumors display a wide variety of clinical and pathological features. The rapid development of molecular technologies over the last decade, using micro-array and expression studies to assess human diseases, has revolutionized the field of breast cancer characterization. The technique is based on simultaneous assessment of thousands of genes used to define the metabolic status of tumor cells and their active interaction with the microenvironment. Using hierarchal clustering to profile tumors according to expression of intrinsic genes, locally advanced breast cancers were classified into several major sub-groups, including: 1. Luminal-epithelial group (with sub-classification into types A and B), characterized by expression of the estrogen receptor and genes associated with the estrogenic function; 2. Basal-epithelial group, typically negative for the estrogen and progesterone receptors and the Her2/neu oncogene; 3. ERBB2+ group associated with over expression of the Her2 amplicone; 4. Normal breast-like group. Those sub-groups are the results of different pathophysiological processes, and their clinical outcomes are unique and predictable. Using supervised analysis, in which clinical outcome is pre-defined, several "poor prognostic expression signatures" were found to be associated with an increased risk of relapse, including: a 70-gene cluster in young patients with node-negative tumors, expression signature characteristic of the wound-response activity in node-positive and/or negative tumors, and the Oncotype DX system relevant for risk stratification and systemic treatment recommendations in node-negative receptor-positive tumors. Benefits and limitations of those methods and future directions are discussed.