A roller bearing fault diagnosis method based on empirical mode deco- mposition(EMD) and sphere-structured support vector machine（SVM）is put forward. The method took the EMD energy feature vectors as the input Parameters to establish the sphere-structured SVM identifying the roller bearing fault pattern, every clustering was visualized for veri fying it by the chromaticity diagram of Maxwell’s triangle plane coordinate．Compared with the method of wavelet packet energy feature vector establishing the sphere-structured SVM,the method of EMD energy for pre-processing can extract fault feature vector more exactly and identify fault pattern accurately and effectively.