Abstract: Strong fluctuation in groundwater level are often caused by typhoon rainstorm, which indirectly affects the stability of geological bodies, and was the primary cause of landslides in southeastern coast area. Therefore, an accurate prediction of groundwater level under rainfall was of critical significance to prevent and early warn of this type of landslides. RBF neural network, that could infinitely approximate any nonlinear function value with AI analysis of sample data, was suitable for dynamically predicting landslides’ groundwater level. Based on the long-term monitoring data of Zhonglin landslide, such as displacement, rainfall and groundwater level, this paper analyzed the seepage and deformation characteristics of typhoon rainstorm-type landslide, discussed the corresponding relation between rainfall and groundwater level. The width of radial base was determined by MATLAB software training, and a dynamic prediction model of groundwater level was established thereby. Then, through the comparison between measured and predicted values of the groundwater level, it was concluded that a minimum deviation value between measured and predicted was 0.01 m, a maximum value 3.13 m, and a mean value 0.46 m. In addition, the more the number of samples at the same rainfall level was, the more accurate the predicted result would be. The research showed that RBF neural network was of practical significance in groundwater level prediction.