The identification of new biological markers for breast cancer has adopted a new dimension by the use of novel techniques such as global gene expression profiling. While important results have been achieved by these methods not all hopes for a more precise assessment of patients' prognosis have yet been accomplished and validation of prognostic or predictive gene signatures is still often difficult. Several recent approaches suggest that comparisons of differential gene expression could be more instructive if prior classifications of tumors based on molecular or biological characteristics were applied. We previously reported a subtype of breast cancer by using a cluster of coordinately expressed genes many of which has been associated with the mammary epithelial stem cells. While a stringent inverse link of ER status and proliferation of the tumor was observed among those "stem cell like" (SCL) tumors, this link was "uncoupled" in about half of the Non-"stem cell like" (Non-SCL) tumors. This subgroup of SCL tumors can be used as a reference system to analyze changes in the ER pathway by comparing the expression of genes dependent on the ER status. By using this strategy we identified Plexin B1, a cell-surface receptor for the semaphorin Sema4D, whose expression is reduced in the group of "uncoupled" tumors. Loss of Plexin B1 is associated with a poor prognosis in both univariate (all patients: p=0.0062; ER positive: p=0.0107) and multivariate analyses (all patients: p=0.032; ER positive: p=0.022). In conclusion those strategies of gene expression analysis in a context of biological meaningful classifications could be helpful to reveal new prognostic/predictive markers.