In this paper we introduce the design and implementation of a next generation short-term flood prediction system for the Three Gorges project. This prediction system is able to take multiple sources of rainfall data as inputs, and applies a distributed geomorphology-based hydrological model (GBHM) to simulation of rainfall-runoff processes. It is believed that by taking account the heterogeneity of land-use and soil types into the hill-slope hydrological processes, the GBHM performs better in reproducing spatiotemporal patterns of hydrologic variables. For the flood routing in the Three Gorges reservoir, MIKE11 model is coupled to simulate the hydrodynamic processes in the Yangtze mainstream across the reservoir. The evaluation results based on historical data shows that this new generation prediction system can capture the hydrological responses accurately, including the floods generated in the Three Gorges region, and the floods routing in the reservoir.