Taiwan is threatened by the disastrous storms that result from tropical cyclones or typhoon every year. The torrential rainfall of these storms, together with Taiwan's special geographical features such as the steep slope and short length of its river systems, often result in the disastrous losses around US$28.03 million. The Lan-Yang Creek basin is located in the low-lying Lan-Yang Plain, where the inundation disasters occur frequently in summers and falls because of the torrential rain. If the real-time river stage forecasting information is available before flooding, and employed for preparation, the damages would be effectively mitigated. The purpose of this study is to develop a Rainfall-Stage Forecasting model applying to the Lan-Yang Creek. This study presents the application of Adaptive-Network-Based Fuzzy Inference System (ANFIS) to construct Lan-Yang Creek flood forecasting model, which allowed nonlinear inputs and outputs. So the study can make consultation for the flood prevention. The study first constructs the fuzzy rules and membership functions. Then the study uses rainfall records of Tuu-ChangANan-Shan and Tay-Pyng mountain and stage records of Jia-Yuan bridge to find optimal distribution of membership functions by processes of the hybrid-learning algorithm. Finally, the study used ten typhoon events including typhoon Masta to forecast five typhoons' stage including typhoon Mindulle. The simulation results show that this model has good accuracy for real-timeA1-hour A2-hour stage and 3-hour river stage forecasting. The root mean square errors of simulation results are between 0.0155 and 0.3190. The new approach of rainfall-stage forecasting model using the ANFIS with rainfall and stages for Lan-Yang Creek really provides a useful tool for flood forecasting.