The status of estrogen receptor (ER) has been profoundly associated with breast cancer. Numerous studies have been conducted to identify informative genes that are associated to ER status. However, the integrity of the reported genes is still inconclusive as the results are derived from small cohort of breast cancer patients (<; 200 samples). In this paper, we studied gene signatures from a cohort of 278 breast cancer samples, labelled in ER positive and ER negative classes, using artificial neural network (ANN). Our model has showed its efficacy for selecting significant genes compared to the previous study. The result also showed that the highly ranked genes have been previously reported in association to the breast cancer development.