Abstract: In the hematite ore roasting process of a shaft furnace, the product quality index, namely the magnetic tube recovery rate (MTRR), is difficult to be measured on-line or in real time. Therefore, a quality monitoring system is developed on the basis of RBF neural network and expert system, involving a MTRR prediction model and a product quality diagnosis model. Practical applications show that the proposed system can timely predict the MTRR with great precision and well diagnose the product quality. Furthermore, the way to adjust relevant parameters is suggested for the process to avoid the unqualified products. As a result, the MTRR is increased by 2% with the qualified product improved 50%, thus the product quality of the roasting process of shaft furnace can be efficiently guaranteed.