The natural progression of breast cancer differs greatly between patients; the precise prediction of this disease course will improve the efficacy of therapeutics. Gene expression profiling may elucidate the undiscovered biological variations between seemingly similar cancers, leading to a new cancer classification system valuable in accurate diagnosis. The expression levels of 2412 genes, derived from 98 cancer samples, were precisely recorded by a high throughput RT-PCR technique, adapter-tagged competitive PCR. Subsequent cluster analysis revealed a molecular profile, correlating with estrogen receptor levels and the presence of lymph node metastases. We analyzed 301 cancer samples for the expression patterns of 21 genes critical in this categorization. The classification of the samples into three major groups was verified utilizing principal component analysis. This molecular classification system correlated significantly with early recurrence, independent of lymph node status. This malignant potential is associated with the expression levels of a group of genes, which comprise a set of candidates potentially useful in diagnostic prediction. These genes and the associated control mechanisms may also be effective therapeutic targets.