Abstract: Celestial navigation method is a kind of important autonomous navigation method for deep space probing. Now it is widely used in many situations, especially in deep space exploration for its special characteristics. There are two kinds of celestial navigation methods. One is to use the geometric technique, in which the navigation information is obtained from the geometric relation between the celestial bodies and explorer. The other is an optimal estimate technique, which uses the celestial measurement as observation directly and employs the extended Kalman filter to estimate the position of explorer. A new autonomous celestial navigation for lunar explorer is developed and investigated in this paper, which combines both these two techniques. At first using the geometric technique to calculate the position of satellite and then reestimate the original position determination results using orbit dynamical equations and Multiple Model (MM) Kalman filter to get a better navigation result. The simulation result shows the high precision of this new method. Finally, the superiority and effectiveness based on the observability and the degree of observability analysis of this method are demonstrated by the simulation result simultaneously.