Background: Several microarray studies have shown that breast tumours can be grouped in at least 4 to 5 individual subtypes namely basal-like, erbB2-like and luminal-like A, B, C or 1, 2, 3. However, although the basal and the erbB2 subtypes are repeatedly recognized as distinct entities, the definition of luminal subtypes has been far from consistent between published series. Refinement of their molecular definition is therefore needed. In this study we sought to better characterize luminal subtypes by applying gene expression grade index (GGI) as previously defined by our group (Sotiriou et al. Proc Am Soc Clin Oncol, 2005), and to correlate them with clinical outcome.Material and Methods: GGI was defined based on the 128 most significant genes identified by gene expression profiling (GEP) in our training set of 64 ER-positive histological grade (HG) 1 or 3 breast tumours. ER-positive tumours were then classified as genomic grade (GG) 1 (GG1, low grade) or GG3 (high grade) based on their GGI values in our unpublished dataset including a series of 154 untreated and 175 tamoxifen-treated breast cancer patients. We also used as external validation sets several previously published datasets from Sorlie et al. 2001, van de Vijver, M. J. et al. 2002, Sotiriou et al. 2003, and Wang et al. 2005. Results: Almost all ER-positive tumours previously defined as luminal-like A or 1 subtype, which had the best clinical outcome, showed low GGI values (low grade). In contrast, luminal B/C or 2/3 ER-positive tumours showed high GGI values (high grade) similar to those of basal-like and erbB2- like subtypes which are predominantly ER-negative. In other words, GGI divided ER-positive tumours into two distinct subgroups namely GG1 and GG3 having statistically distinct clinical outcome in either untreated (HR: 3.93, 95% CI 2.28–6.76, p