This work presents a method for automatic mass segmentation located in regions of interest of mammographic images. The segmentation is made using the technique of region growing. Its threshold parameter is automatic defined using a decision tree, constructed by supervised training using the J4.8 algorithm, which computes attributes based on image statistics. The system was developed using the open source tools ImageJ and Weka. Both real and phantom images were used in the tests. The tests result in 79% of correct segmentations of the regions of interest.