A new fault diagnosis model is proposed based on Littlewood-Paley wavelet support vector machine(LPWSVM): firstly, fault vibration signals are decomposed into several stationary IMFs, then the instantaneous amplitude Shannon entropy of the IMFs which have fault modulation characteristics is computed and regarded as the input characteristic vector of LPWSVM for fault classification. EMD decomposition adaptively isolates fault modulation signals from original signals. The distinction among instantaneous amplitude Shannon entropy vectors enlarge the differences among fault types. Littlewood-Paley wavelet kernel is a new multivariable support vector kernel function,which has orthonormal feature, and can approximate almost any function in quadratic continuous integral space. Thus it has favorable nonlinear mapping ability. Compared with LS-SVM under the same conditions, the learning precision and adaptive diagnosis capacity are improved by LPWSVM. Therefore,it will be more suitable for complicated pattern recognition. The rolling-bearings fault diagnosis example proves the effectivity of this new model.