Automatic detection and classification of lesions in mammography remains one of the most important and challenging problems in the development of computer-aided diagnosis systems. Several machine learning approaches have been proposed for supporting the detection and classification of mammographic findings, and are used as computational tools during different diagnosis process by the radiologists. However, the effectiveness of these approaches depends on the accuracy of the feature representation and classification techniques. In this paper, a radiomic strategy based on texture features is explored for identifying abnormalities in mammographies. For doing that, a complete study of five feature extraction approaches, ten selection methods, and five classification models was carried out for identifying findings contained in regions of interest extracted from mammography. The proposed strategy starts with a region extraction process. Some square regions of interest (ROI) were manually extracted from the Mammographic Image Analysis Society (miniMIAS) database. Then, each ROI was decomposed into different resolution levels by using a Wavelet transform approach, and a set of radiomic features based on texture information was computed. Finally, feature selection algorithms and machine learning models were applied to decide whether the ROI undergoing analysis contains or not a mammographic abnormality. The obtained results showed that radiomic texture descriptors extracted from wavelet detail coefficients improved the performance obtained by radiomic features extracted from the original image.