In order to solve the difficulty in fault feature extraction of the planetary gearbox vibration signal,a comprehensive fault diagnosis method based on empirical mode decomposition（ EMD）,singularity value decomposition（ SVD） and probabilistic neural networks（ PNN） is proposed. Firstly,with the EMD method,de-noised vibration signals are decomposed into a finite number of intrinsic mode functions （IMF）. Secondly,some IMF components are selected according to the criterion of correlation coefficient and variance contribution ratio,and singular value sequences regarded as eigenvectors are obtained with the method of SVD. Lastly,the eigenvectors serve as import of PNN so that faults of the planetary gearbox are recognized. Experiments are conducted on the planetary gearbox fault diagnosis test-bed and comparison is carried out with eigenvectors made of energy entropy,which fully validates the effectiveness of the proposed methodology.