Abstract: Passive intermodulation is one of the key factors which have an influence on the performance of communication systems. It is found that the PIM signal is not a fixed value, but a non-stationary signal which varies with time. With the characteristic of time series of PIM signal, a predicting method on PIM power is proposed based on wavelet neural network in this paper. Firstly, the predicting model and its predicting method on wavelet neural network are described in detail. Secondly, as the verification object, the coaxial connector is chosen and the time series of third-order PIM power is obtained through the PIM testing system. Lastly, predicting model based on the wavelet neural network is used to predict the subsequent PIM power with the obtained time series. And it is verified that the wavelet neural network has a good performance on predicting PIM power by comparing the predicted results and experimental results. This study is helpful to develop suppression technology on PIM.