A method of rolling bearing fault diagnosis using the wavelet packet t ransform and the least square support vector machine (LSSVM) was proposed accor ding to characteristics of the bearing vibration signal. The reconstructed signa l of each frequency ranges and the energy of each decomposed node was obtained b y decomposing the bearing vibration signal using the wavelet packet. The energy of each node was regarded as the eigenvector of diagnosis models and was input to the LSSVM multiclassifier to recognize failures. A test was conducted and the vibration signal was measured from a rolling bearing experiment table. The res ult shows that the method has a faster convergence speed and a higher degree of classification precision than conventional methods.