The detection of partial discharge (PD) signal is generally noised. It is bad for the recognition rate and velocity of partial discharge recognition. The recognizer based on moment feature and probabilistic neural network (PNN) is proposed to improve the recognition rate and velocity of partial discharge recognition. The moment feature is extracted from the three-dimensional spectra of PD signal. The moment feature is regarded as the input of PNN. Then, the partial discharge recognition is done. The proposed recognizer is compared to Back Propagation (BP) neural network, Bayesian classifier and extreme learning machine (ELM) on the needle-plane discharge and the ball-plane discharge. Experiment results show that the recognition of the proposed recognizer based on moment feature and PNN is satisfactory.