Breast cancer is a heterogeneous disease, encompassing a plethora of histological types and clinical courses. Current histopathological classification systems for breast cancer are based on descriptive entities that are of prognostic significance. Few prognostic markers beyond those offered by histopathological analysis are available. Furthermore, a very limited armamentarium of predictive biomarkers has been introduced in clinical practice. High throughput molecular technologies are reshaping our understanding of breast cancer, of which microarray-based gene expression has received the most attention. This method has been successfully used to derive a molecular taxonomy for breast cancer, which has provided interesting insights into the biology of the disease. Microarray-based class prediction studies have generated a multitude of prognostic/predictive signatures. Although these signatures have not been fully translated to clinical practice as yet, they herald the promise of an improvement in breast cancer treatment decision-making. It should be noted, however, that most of the signatures developed to date seem to have discriminatory power almost restricted to oestrogen receptor-positive disease. This review addresses the contribution of gene expression profiling to our understanding of breast cancer and its clinical management and what has yet to be done for these classifiers to be incorporated in clinical practice.