The optimal Kalman filtering of GNSS/SINS integrated navigation system needs to satisfy the assumption that the measurement vector is Gaussian distribution; otherwise, the filtering accuracy will decrease or even diverge. Robust filtering is an effective method to solve this problem, but it has certain limitations in improving the filtering accuracy. To solve this problem, this paper proposes an improved robust filtering method based on Mahalanobis distance and Newton iteration method. Firstly, the Mahalanobis distance is used to determine whether the measured value is abnormal. Secondly, a scaling factor is introduced to optimize and adjust the measurement noise covariance matrix according to Newton iteration method, and then adjust more precisely the proportion of measurement value and system model information in the filtering process. Finally, the GNSS/SINS integrated navigation system is simulated and analyzed based on optimal Kalman filter (KF), traditional robust Kalman filter (RKF) and improved robust Kalman filter (IRKF). The experimental results show that IRKF can improve the position and velocity accuracy by about 14.1% and 13.8% respectively when compared with KF, and can improve position and velocity accuracy by 8.1% and 7.7% respectively when compared with RKF.