Abstract. The accumulated error of GLONASS broadcast ephemeris is nearly 1 meters within 15 minutes due to simplified perturbation force model. Aiming at this problem, a new method of GLONASS broadcast ephemeris fitting algorithm is introduced. In this method, a nonlinear measurement model is built based on the GLONASS broadcast ephemeris user algorithm and linearized. And the partial derivative of Jacobin matrix is computed by difference quotient approximation. Then the GLONASS broadcast ephemeris parameters are estimated by using least square method. The simulation results show that the average user range error (URE) caused by fitting broadcast ephemeris is improved about two orders compared with the traditional broadcast ephemeris and achieves millimeter level. Finally, a verification system was built by the Japanese RTKLIB tools, the availability of the fitting broadcast ephemeris is validated from the positioning angle.