Abstract: To solve the problem that the pure linear or nonlinear filter is difficult to obtain the optimal state estimation results of a hybrid linear/nonlinear dynamic system, a novel hybrid dynamic filter based on Kalman filter(KF) and Augmented Cubature KF(A-CKF) was proposed by utilizing the advantages that KF can obtain optimal solution for linear state and its computation cost is low. In the presented filter, the system state was decomposed into linear and nonlinear parts which were estimated by the KF and the Simplified Twice Augmented Cubature KF(STA -CKF) respectively. The simulation results of the maneuvering target tracking and the strapdown inertial navigation system nonlinear initial alignment cases show that the novel filter had similar filtering accuracy to the Rao-Blackwellized particle filter but lower computation cost. Compared with the pure STA-CKF, its accuracy and real-time performance were improved significantly.