As there have been many widely used watershed hydrological models, it now becomes possible to select a number of different hydrological models in parallel running for rainfall-runoff simulation. Under the condition of the same rainfall inputs, however, different models will produce various runoff estimates, and the model efficiency and accuracy will also differ from each other. Thus, the very crucial problem is how to combine the simulation results from a number of different simultaneously running hydrological models for improving the model efficiency and estimate accuracy. In this paper, the adaptive-network-based fuzzy inference system (ANFIS) is chosen as a combination framework, which has been applied to combine the simulation results of two hydrological models, one is the three-source Xinaujiang model and the other is the simple linear model (SLM). The combination results have been demonstrated more robust and accurate than those of either of the two hydrological models. It is recommended that ANFIS be used in practical runoff simulation and forecasting.