Multistage rocket boost phase tracking mainly adopted decoupling Kalman filtering method based on current statistical model, when the maneuvering parameters of the current statistical model could not be adaptive to changes, separating section would take on filter kicking. To solve this problem, an adaptive calculation of model parameters was put forward and the cubature Kalman filter was used to solve nonlinear filtering problem. Through the analysis of measured data of a certain rocket flight, the algorithm could effectively restrain filter kicking, and the stability and convergence of the algorithm were obviously better than traditional scheme, and it was suitable for multistage rocket boost phase tracking.