Kalman filtering technique can be used as optimal estimation of motion state of a LEO (Low-Earth-Orbiter); it can be applied directly to real time Orbit Determination (OD), or to Precise Orbit Determination (POD) with post-processing mode. The OD accuracy using Kalman filters depends on a priori knowledge of system models, noise statistics, especially quality of observations from onboard GPS receiver on LEO. When using Kalman filtering technique for GPS-based OD of LEOs, the gross errors in observation equation must be well dealt with firstly. Having analyzed the characteristics of solving gross errors in pervious methods, QUasi-Accurate Detection (QUAD) of gross errors method has been employed to detect and correct gross errors in observations. The advantages of this method are:accuracy rate of detecting gross errors is high; multiple gross errors can be detected at the same time. In addition, UD decomposition technique and Sage adaptive filter are employed to overcome the instability of numerical value computation and divergence of filter that properly occurred. At last, a simulation example of CHAMP satellite is used to demonstrate the feasibility and validity of new method presented in this paper.