We have developed a new algorithm for the characterization of microcalcification clusters. Fuzzy logic is well suited to represent and to manipulate data and knowledge at different levels of the algorithm. Our algorithm is built in 3 steps: Detection and segmentation of the individual microcalcifications, measurements on the segmented microcalcifications (shape, contrast, relative localization), use of these measurements as inputs of a learning system which concludes if the current case is malignant or not. We first describe some aspects of the fuzzy segmentation we have implemented. Then we explain how we build fuzzy measurements from the segmented objects and how these measurements are manipulated into the fuzzy decision tree we are using. Finally, we present the preliminary results we obtained with our test database.