Abstract: A new cascade nonlinear inferential control strategy for a nonadiabatic fixed bed reactor based on improved hybrid model is proposed. Firstly a new inferential estimator of the fixed bed reactor for phthalic anhydride synthesis based on the improved hybrid model is developed. Then a new nonlinear adaptive control algorithm is employed to design secondary controller, improved program variable gain PID is proposed to design the primary controller, and a new nonlinear filter is designed to overcome the difficulty that primary process is much faster than secondary process. Finally the cascade nonlinear inferential control system based on the improved hybrid model is designed. Simulation results show that the proposed estimator possesses the advantages of high estimation accuracy, good real time nature, being accurate to interpolate and extrapolate and easier to analyze and interpret, and requiring significantly fewer training examples, and that the cascade nonlinear inferential control system provides excellent tracking performance and adequate disturbance rejection performance, substantively improves system reliability and control performance, and is much better than single loop inferential control system.