The invention provides compositions and methods for characterizing breast cancer stem In particular, the invention provides for the identification of cells expressing Twist and CD44 that express little or virtually undetectable levels of CD24 (i.e. a Twist+/CD44+/CD24−/low cell sub-population). The presence of such cells in a breast cancer specimen identifies the breast cancer as having increased metastic potential. Such cancers are identified as requiring aggressive therapies. Accordingly, the invention provides biomarkers suitable for identifying, diagnosing, and monitoring treatment of a subject with breast cancer.