Breast cancer is one of the most deadly related diseases in women across the world. The survival rate among the patients with the breast cancer will increase, if the disease is detected earlier. Mammogram analysis is one of the most promising methods that are being used in the early detection and abnormality classification of the breast cancer. Irrelevant and noisy features extracted from mammogram image often mislead the learning processes and also have negative impact on the quality of classification process. Therefore, this paper proposed the use of Fuzzy Rough Set Method to select the most significant texture features from mammogram images. Selected features are employed to build a more easy and understandable learning model in order to improve the classification quality of mammogram analysis systems. The results show that the proposed method selects the appropriate subset of features that are mostly representing the original data and increase the quality of classification. © 2013 IEEE.