In order to improve the efficiency and accuracy of fault diagnosis， a fault diagnosis method of rolling bearing of the adaptive no velocity term particle swarm optimization-back propagation neural network （ANVTPSO-BPNN） with optimal wavelet packet energy （WPE） is proposed. The wavelet analysis is used to denoise the vibration signal， and the energy feature is extracted by the wavelet packet decomposition. The basis function and decomposition level are optimized. The inertia weight and learning factor of particle swarm optimization （PSO） algorithm are adaptively adjusted， and the speed term of standard PSO is discarded to avoid the influence of particle initial velocity on the convergence speed and solution accuracy. According to the measured data of a certain bearing， five different BPNN algorithms are verified. The results show that the proposed method has only 273 iteration steps and its diagnosis accuracy reaches 99%. Compared with BPNN before and after noise elimination and two PSO-BPNNs after noise elimination， the method has the higher diagnosis efficiency and accuracy.