Abstract: A coupled neuro network model based on discrete wavelet transform and long short-term memory is introduced into the water level prediction. The model is verified with the data from The Hankou Hydrological Station. The models of typical BP neuro network, DWT-BP neuro network. LSTM neuro network are compared to the DWT-LSTM coupled neuro network model. It is found that the all of the 4 types of models can achieve the water level prediction as accurate as 90%, but for rapid change of water level, BP model gives obvious error. In such a case the DWT-LSTM coupled neuro network model is superior, 10.9%（1 or 2 days in advance） or 25.2%（4 or 5 days in advance） better than LSTM model.