The invention relates to a forecasting method of the water level of potamic tidewater. A traditional forecasting method of tidal hour comprises a time propagation method and a time isolation retardation method which weaken the influence of multiple factors, such as tides, radial flows, riverway terrain, wind power, wind directions, and the like on the tidewater to a certain extent. The forecasting method of the water level of the potamic tidewater comprises the following concrete steps: firstly, constructing a neural network model based on a neural network function in a kit function library of MATLAB 6.5 and the history data of tidewater water levels and time; then, utilizing a neural network training function in the kit function library to train a network; utilizing a simulation function to test the network; and finally, using the trained and tested neural network model to forecast a water level value of the next high tide level or the next low tide level. The invention uses the history tidewater data to forecast the water level value of short-term tidewater and can fully neglect the influence of uncertain factors, such as wind directions, rainfall, water supply and drainage, river-bed variation, and the like.