Abstract: Xiangyang coal with low ash fusion temperature (AFT) and Jincheng coal with high AFT were used to prepare the blending samples. The influence of Xiangyang coal addition on AFT of Jincheng coal was explored by XRF, SEM, DSC, XRD, and ternary phase diagram analysis. The results show that blending coal can reduce the AFT effectively. The AFT of blending coal is lowered significantly when the adding amount of Xiangyang coal is lower than 24%. Whereas, when the adding amount is between 24% and 40%, AFT of the mixed coal has a slight change and the ash flow temperature is below 1 400℃. A series of chemical reactions among ash composition of mixed coal occur at 1 000-1 200℃, mainly including formation of high melting point compound (mullite) from SiO₂ with A1₂O₃, and that of low melting point compounds (anorthite and hercynite) from the reactions between mullite and CaO or Fe₂O₃. The above reactions mainly cause the changes of ash fusion temperature in blending coal. Based on BP neural network, a prediction model of ash fusion temperature was built. It is proved that the prediction average accuracy by BP neural network is higher than 99%, which is better than that of a previous empirical formula. Furthermore, analysis by thermodynamics software (HSC 5.0) shows that mullite prefers to react with CaO rather than Fe₂O₃.