Abstract: Exhaust gas temperature is one of the performance parameters which reflect aeroengines- running state most efficiently. The prediction analysis of the sequent takeoff exhaust gas temperature margin (EGTM) is helpful to estimate aeroengines- future working performance, which can offer sufficient time reference and decision-making support for the fault prevention and elimination. When building the prediction model according to the EGTM historical observation sequence which was characterized by nonlinearity and nonstationarity, a solution combining radial basis function prediction networks (RBFPN) and functional coefficient autoregressive model (FAR) was proposed based on the sequence partition with singular value decomposition filtering algorithm. The respective advantages of RBFPN and FAR in modeling the trend element and the random element of EGTM sequence were taken complementally and cooperatively. It is indicated by experimentation that the solution can effectively restrain the shortcomings of separate employment of RBFPN or FAR, and improve the prediction performance.