Abstract: The celestial navigation detects the celestial bodies through measuring instruments and calculates the navigation information of the carrier. It has the characteristics of strong concealment and no accumulation of errors over time. However, the traditional celestial navigation is heavily dependent on the horizontal attitude of the inertial measurement unit(IMU) and the navigation precision is limited. Based on the characteristics of satellite/stars, this paper presents a celestial navigation method based on optical observations of satellite/star. The star sensor is used to simultaneously image the satellite and stars. Taking the star vector as the relative reference, the observation vector of the satellite is calculated by the least square method. Combined with the satellite's ephemeris orbit prediction position, the line of position is determined. Through multiple observation vectors, the navigation information of the carrier is solved by the Newton iterative method. The influences of different types of error sources of this method on the accuracy of navigation solution are analyzed respectively from theoretical and simulation perspectives. The simulation results show that when tracking and observing low-orbit satellites with a satellite pointing measurement accuracy of 0.02º, the navigation accuracy is better than 100 m. Under the condition that the satellite's pointing accuracy is 1′, four satellites in a 300 km orbit are observed, and the navigation accuracy is better than 100 m. When the number of observed satellites increased to 8, the navigation accuracy improved by 35%. The method proposed in this paper enables autonomous position determination, offering a novel approach for the development of next-generation celestial navigation equipment, and holds potential engineering application value in the context of large-scale satellite constellation deployment.