Abstract: Based on classification of silica and alumina content and silica alumina ratio in coal ash, optimized prediction models of coal ash melting point and viscosity related to ash components were built. Their relationship in a wide range of ash components was also discussed. An ash fusion temperature prediction model based on liquidus temperature from FactSage Thermodynamic software demonstrates accuracy improvement with error of less than ±40 ℃ and deviation of 25 ℃. The combination of Urbain and Roscoe model results in viscosity prediction error between prediction and experiment results within ±0.1 for relative low viscosity and ±0.2 for relative high viscosity. The results show that components classification model obtains better results than wide-covered model. Thus a better understanding of ash fusion mechanism is provided and gives reasonable guidance for blending and additives added to coal.