Abstract: Considering the difficulty of on-line measurement of matte grade in copper flash smelting process and based on analysis of the components,the independent chemical reaction and the molar relationship among components are studied and a mathematical model is presented.Due to the modeling simplifications and complexity of the reaction mechanism,the matte grade prediction precision of the mathematical model can not satisfy the needs of practical applications.Then based on industrial running data,a neural network predictive model for matte grade is established,which can satisfactorily describe the relationship among the training sample data but has low generalization ability.In order to overcome the limitations of single model,a fuzzy coordinator with an adaptively adjustable membership function is introduced,and an intelligent integrated prediction model is proposed,which integrates mathematical model with neural network model.Industrial running results validate the effectiveness of the presented model.