A filtering algorithm, cubature Kalman filter-Kalman(CKF-KF) filter, is proposed for conditionally linear Gaussian state model, which respectively employs CKF and KF to estimate nonlinear state and linear state in the model. The above states are carried out cubature sampling, which are propagated through linear and observation equations to estimate nonlinear state. The maneuvering target tracking simulation results show that, compared to the Rao-Blackwellized particle filter(RBPF), the algorithm running time of CKF-KF is less than 1% of that with a slightly lower filtering performance loss, and the estimation accuracy of CKF-KF coincides with that of UKF-KF, whereas the algorithm running time reduces by 22% and effectively improves real-time.