Purpose: To assess the ability of genes selected from those reflecting chromosomal instability to identify good and poor prognostic subsets of Grade 2 breast carcinomas. Methods: We selected genes for splitting grade 2 tumours into low and high grade type groups by using public databases. Patients were diagnosed between 1999–2002 at the Budai MA´ V Hospital. 187 formalinfixed, paraffin-embedded breast cancer samples were included in the qPCR-based measurement of expression of AURKA, FOXM1, TOP2A and TPX2 genes. The expression of the genes were correlated to recurrencefree survival (RFS) and immunophenotypical characterization of tumours. 1509 samples were in silico analyzed for further validation of the selected genes. Results: Grade 1 and 3 groups were used as training set for the selected genes. The 4-gene signature was able to split grade 2 carcinomas (n = 62) into a good and a poor prognosis group (RFS: 83.8±4.9 months and 69.4±8.2 months, respectively, p = 0.016). Furthermore, independent of grade, the identified signature containing only TOP2A and FOXM1 (TOPFOX) was able to separate ER+ tumours in an efficient manner (p = 0.009), which is further supported by validation in a dataset containing 1509 patients (p = 8.1E-8). Conclusions: The selected genes with the appropriately selected control genes are able to separate the different prognostic subgroups independently from histological grade. Our results show the feasibility of the selection of a minimal set of genes for the development of a clinically applicable prognostic test. The work was partly supported by the TA´MOP.4.2.1.B-09/1/KMR-2010– 0001 grant.