Abstract: Grassland fire disaster is one of a sudden, destructive, and hard to disposal grassland disasters, it caused a great threat to the pastoral people’s lives and property in Hulunbeier grassland fire statistics on statements and related meteorological data from 1994 to 2005 and theories of risk of natural disasters, and of regional disaster system, an astoral. In this research, early warning of grassland fire disaster risk was modeled through the study of six livestock counties of Hulunbeier grassland, based on data of disaster early warning. Logistic regression model was used in the identification of the key factors influence the early warning of grassland fire disaster risk; in the calculation the weight of individual indicators, analytic hierarchy through scoring by experts was used; grid GIS technology combined with regression analysis was used in the spatial distribution of indicators of the constituency which will make space rating scale more accurate; warning source of endogenous and exogenous sources of warning signs of Hulunbeier grassland fires was analysis through weighted comprehensive analysis; with the optimal partitioning of the fire cases from 1994 to 2004, early warning threshold was identified and was divided into the blue, yellow, orange and red alert. Grassland fire risk warning has been calculated on Hulunbeier, take all the fires on Hulunbeier grassland in 2005 for instance. It shows a good agreement between high risk fire areas of early warning and fire point, which, in terms, tested the accuracy of the model in a certain degree.