Abstract: Traditional hydrological prediction is a type of deterministic prediction, which does not make full use of the weather forecast information, or give forecast uncertainty. This paper analyzed the precipitation hindcast data provided by the Global Forecast System (GFS) in National Centers for Environmental Prediction (NCEP); and then the hindcast data was processed by the ensemble pre-processor in order to drive the hydrological model. Finally, we got the hydrological ensemble prediction. The results show that the hydrological ensemble prediction is better than the traditional ensemble streamflow prediction (ESP). The hydrological ensemble prediction on GFS is an effective way of developing hydrological prediction.