INTRODUCTION: Few markers are available that can predict response to tamoxifen treatment in estrogen receptor (ER)-positive breast cancers. Identification of such markers would be clinically useful. We attempted to identify molecular markers associated with tamoxifen failure in breast cancer. METHODS: Eighteen initially ER-positive patients treated with tamoxifen requiring salvage surgery (tamoxifen failure [TF] patients) were compared with 17 patients who were disease free 5 years after surgery plus tamoxifen adjuvant therapy (control patients). cDNA microarray, real-time quantitative PCR, and immunohistochemistry on tissue microarrays were used to generate and confirm a gene signature associated with tamoxifen failure. An independent series of 33 breast tumor samples from patients who relapsed (n = 14) or did not relapse (n = 19) under tamoxifen treatment from a different geographic location was subsequently used to explore the gene expression signature identified. RESULTS: Using a screening set of 18 tumor samples (from eight control patients and 10 TF patients), a 47-gene signature discriminating between TF and control samples was identified using cDNA arrays. In addition to ESR1/ERalpha, the top-ranked genes selected by statistical cross-analyses were MET, FOS, SNCG, IGFBP4, and BCL2, which were subsequently validated in a larger set of tumor samples (from 17 control patients and 18 TF patients). Confirmation at the protein level by tissue microarray immunohistochemistry was observed for ER-alpha, gamma-synuclein, and insulin-like growth factor binding protein 4 proteins in the 35 original samples. In an independent series of breast tumor samples (19 nonrelapsing and 14 relapsing), reduced expression of ESR1/ERalpha, IGFBP4, SNCG, BCL2, and FOS was observed in the relapsing group and was associated with a shorter overall survival. Low mRNA expression levels of ESR1/ERalpha, BCL2, and FOS were also associated with a shorter relapse-free survival (RFS). Using a Cox multivariate regression analysis, we identified BCL2 and FOS as independent prognostic markers associated with RFS. Finally, the BCL2/FOS signature was demonstrated to have more accurate prognostic value for RFS than ESR1/ERalpha alone (likelihood ratio test). CONCLUSIONS: We identified molecular markers including a BCL2/FOS signature associated with tamoxifen failure; these markers may have clinical potential in the management of ER-positive breast cancer.