The strap-down inertial/stellar refraction integrated navigation method (SINS/RCNS) was proposed in this paper based on high precision of starlight refraction indirect sensitive horizon. The outputs posture of the star sensor correct gyro drift, and combined with the atmospheric refraction model, the starlight refraction angle can be got which is used to correct the position and velocity error. The observation model of the integrated navigation system was derivated, and it's observability was analyzed, finally the Kalman filter (KF) was used for state estimation. The simulation result indicated that the proposed method was able not only to estimate the gyro drift exactly, but also to correct the navigation error caused by the accelerometer bias, the divergence of the speed and position error was bated ulteriorly, therefore it was a practical method to improve the positioning accuracy. 