Fire spreading is of an utmost intricate flaming phenomenon, the scientific fire spreading model is the key to forecasting forest fire behaviors. Generally, it is complicated and time-consuming to find improvement and appraise the fire-spreading model, which is applicable to a certain forest area. At present, the method to forecast is a single fire-spreading model or the one selected by commanders. However, based on the Fuzzy Data Mining Technique, indices for fire behavior are established to form fire data warehouse. These are quantity of forest combustible, auriferous quantity of it, inflammability and slope in Guangzhou. So the discovering and forecasting function of Fuzzy Data Mining technique are used on forecasting forest. According to the handy principle, three modes are discovered and the corresponding fire-spreading mode is matched for the certain style, which is induced to form the fire data warehouse. Furthermore, so long as the real-time fire indices are provided by commanders, the close model can be distinguished. Therefore, the suited fire-spreading model can be selected automatically to forecast the forest fire behavior, in this way, dependability of forecasting can be gained.