Abstract: For low Earth satellites, thermospheric density models are widely used in orbit determination and prediction. However, typical models often show density errors of 15 %~30 % under normal condition and the errors may be much greater during unusually solar and geomagnetic period. Density errors can be translated into orbit errors and effect mission plan, re-entry operations and precise orbit determination. In this paper, on the basis of detected data of Shenzhou, a kind of model calibration is expected to be developed on real-time detected data. By analyzing and comparing the character of density error for the national atmosphere model NRLMSISE-00, a kind of average calibration method is developed for relative quiet geomagnetic field (Ap≤30). During the quiet geomagnetic field, the errors of model at the same local time and latitude are considered approximately identical, so the error may be substituted by average error at the same station. The results of model are tested based on Shenzhou 4 data by two means and both are improved in accuracy. The errors of 20 % decreased to 6 % by real-time error calibration. The errors of one day in advance decreased to 7.8 %, two days in advance decreased to 9.4 % and three days in advance decreased to 10.5 %.