In the context of image processing algorithms for mass detection in mammography, texture is a key feature to be used to distinguish abnormal tissue from normal tissue. Recently, a texture model based on a multivariate gaussian mixture was proposed, of which the parameters are learned in an unsupervised way from the pixel intensities of images. The model produces images that are probabilistic maps of texture normality and it was proposed as a visualization aid for diagnostic by clinical experts. In this paper, the usability of the model is studied for automatic mass detection. A segmentation strategy is proposed and evaluated using 79 mammography cases.