Abstract: To solve the problem of water level prediction at ungauged sections, the method of coupled BP neural network and MIKE11 model was proposed in this study. First, the water level, precipitation and antecedent influencing factors were collected for correlation analysis according to the hourly flood events data. Second, the input factors with the highest correlation degree of water level in the Lanjiang River Basin in the prediction period of 12 h and 24 h were selected, the water level predictions at Lanxi station were built by MLR and BPNN models for different forecasting lead times. Then, the BPNN-based water level prediction at ungauged sections in Lanjiang basin was constructed through the synthetic water level data generated by MIKE NAM-HD model. The results indicate that: (1) The performance of BPNN is better than MLR model for water level predictions of long lead times; (2) The NSE obtained from BPNN gradually decreases with the increasing of forecasting lead time; (3) After input variable selection, the improvements of BPNN at lead times of 12 h and 24 h are about 9.0% and 34.7% in terms of NSE; (4) The proposed BPNN coupled with MIKE11 model is able to apply to ungauged section water level predictions.