In order to solve the problem that Kalman filter(KF) used in continuous, non-linear inertial navigation system(INS) suffered low accuracy in estimating model error, model predictive filter(MPF) directly processing continuous, non-linear system was adopted to provide one-step prediction for KF. Combining the advantages of the two filter algorithm, the accuracy of navigation could be further improved. Based on data tested by fiber-optic gyro(FOG) INS and star sensor(SS) data provided by computer, a simulation was executed for the presented combined filters, which proved the feasibility and superiority of the combined filtering algorithm in this paper.