Abstract: A parallel kalman filter system for terrain-aided gurdance based on the theory of Gaussian Sum to meet the requirement of locating flying vehicles with large initial ppsition error is pressented. An order-reduced model is put forward, based on the error analysis of which an adaptire sequential decision （ASD） method is presented. Combining both of then the position uncertainty and the amount of calculation can be decreased in steps and the real-time calculation can be implemented. The performance analysis of ASD gives a practical way to further increase the speed of convergence. It is shown that a fast and accurate location can be provided by the system for a vehicle even with larger errors in position, velocity and course.