Histological grade is a historically used and well-documented prognostic indicator in breast cancer. There are three categories of grade (G1, G2 and G3) based on the degree of tubule formation, nuclear pleomorphism and mitotic count. A number of studies have reported that histological assessment is not uniformly reported. As a result of low inter-pathologist correlation associated with pathological diagnosis and non-standardised grading systems, patients are not always allocated into the correct grouping, and G2 has often been considered a "safe" group if one is unsure. A previously published study used real-time polymerase chain reaction (RT-PCR) for 5 genes to molecularly classify the G2 tumours into either G1 or G3. Due to the workflow constraints within pathology laboratories it was not considered feasible to molecularly profile every G2 tumour. In light of this we obtained the antibodies that corresponded to the 5 genes (BUB1B, CENPA, RACGAP1, RRM2 and NEK2) and performed immunohistochemistry (IHC) on formalin fixed paraffin embedded (FFPE) sections of 43 tumours (11 G1 and 32 G3). Results for all tumours were randomly divided into training and testing sets and an artificial neural network (NeuralSight and NeuralWare Predict) was used to classify the grade of tumours. Thirty-three additional G2 tumours were used for validation of the ANN. The ANN classified these tumours into 5 G1 and 28 G3 tumours. This predicted grade showed significant correlation with patient survival. Neural networks can be used to reclassify breast cancer G2 tumours into G1 and G3 using a panel of 5 IHC markers. This has the potential to impact on patient care, treatment decisions and outcome.