Grassland fire disaster is one of the most destructive grassland disasters, which is sudden, destructive, and hard to handle. It caused a great threat to the humanity and property in pastoral. In this study, a dynamic risk assessment model was built based on the data of prairie fire statistics and related meteorological data from 1994 to 2005 in the six livestock counties of Hulunbuir grassland. Logistic regression model was used in the identification of the key factors influence grassland fire disaster risk. In the calculation of 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 which has higher resolution than counties. Assessment factors of endogenous and exogenous sources of warning signs of grassland fires were analyzed through weighted comprehensive analysis. Fire disasters from 1994 to 2004 were taken as the sample cases to determine the threshold of risk level by using optimal partition method. Taking the grassland fires in 2005 as examples to validate the dynamic risk assessment model, results shown high risk areas coincide well with the distribution of fire points, this proved the accuracy of the model. The dynamic risk assessment results could be used to guide and manage the mitigation of grassland fire disaster and rescue goods distribution.