The shaft furnace roasting process is an important procedure in the mineral processing of the weakly magnetic iron ore. Its technical performance index is called the magnetic tube recovery rate(MTRR), which closely related to the overall performance of the mineral processing. In this paper, we mainly concern on the decision making of the supervisory control of the roasting process to control its MTRR into the target range. This model replaces the human operators to determine the set-points of the lower control loops. The experiment is given to evaluate the proposed model and the results show its validity and efficiency.