Abstract: China has widely distributed coal resources; however, suffering from the perennial underground coal fires for so many decades.The coal fire swallows coal resources and leads to sever environmental and ecological pollution.For fighting against the underground coal fire, the results from statistical analysis of the fire spatial distribution and its probability of the occurrence can be provided to the fire fighters as the theory for reasonably allocating resources and supporting the decision-making.For underground coal fire forecasting and analysis, the data of each factors are with different varieties, and the samples are always not sufficient, binary logistic regression model can overcome these limitations, with its low requirements on the dependent variable data hypothesis, it can be used to predict the dependent variable having two characteristics of probability values.The paper selects the control points generated randomly and the coal fire zones in the database as the sample points, extracts the data in each variable including ash content, pyrite, volatile, calorific value of coal, geological age, faults, slope, aridity, population density above coal seam and the management level of mines.The forecasting model by the binary logistic regression model was established and the model was applied to the north China, the underground coal fires predicting precision can reach to 67.7%, the result can be provided as the theoretical basis for reasonably allocatine fire extinguishing resources.