Aimed at the disadvantages of constant deviation and discontinuity of wavelet soft and hard threshold function and the difficulty in parameter selection of leastsquares support vector machine kernel function, an engine bearing fault diagnosis method was presented based on exponential wavelet threshold and PSO-DP-LSSVM.The original signal was decomposed and reconstructed by using the exponential wavelet threshold function, and the energy feature of each component after noise reduction was extracted. Adaptive DP algorithm was used to enrich the solution space of PSO algorithm, and dynamic parameter control was adopted to get the optimal solution more easily. The energy feature was input into the LSSVM with determined parameters, and the information was trained and predicted. The results show that the fault bearing signals can be rapidly and efficiently diagnosed and classified by using the method.