Abstract: Magnetic Tube Recovery Rate(MTRR) is a key technical index used to evaluate roasted iron mine's quality.The value of MTRR is acquired by chemical test,a process that has a big delay and is hard to measure online after shaft furnace roasting process.To deal with this problem,an intelligent hybrid prediction model is developed to predict the MTRR punctually based on the neural networks,case-based reasoning and expert systems.This model consists of a neural network prediction model,a case-based reasoning prediction model,and a self-tuning model.The whole model's framework,the main function and the algorithm implementation are discussed.Application to shaft furnace roasting process of a Minerals Processing Factory show that the model is effective in both normal and abnormal operating conditions.This prediction model has been applied to the optimal control of the roasting process,so that the MTRR can be kept within the optimal ranges,and obvious benefit is achieved.