Based on the analysis of the mechanism of stray light of the FY‐2 meteorological satellite, the main component of the stray light, generated from the folding mirror direct reflections, is modelled physically. Through the analysis of the stray light outside of the Earth region as well as extracting the high‐order statistical eigenvalues of the designated areas, the stray light function matrix, A, can be calculated using the total least squares (TLS) method, which is applied to the inside of the Earth region, and the full field‐of‐view (FFOV) stray light estimation can be produced efficiently. It is shown that the mean restoration errors of the stray light outside of the Earth region of the infrared (IR), water vapour (WV) and visible (VIS) channels are less than two scaled units, while the visual images are improved greatly. The images of FY‐2B IR with stray light removed and the FY‐2B raw images are inter‐calibrated with the NOAA‐17 Advanced Very High Resolution Radiometer (AVHRR) Ch4 data. The results show that, from 180 to 300 K of the brightness temperature area, the standard deviation of the data with stray light removed is from 1.3 to 3.8 K relative to the NOAA‐17 data, which is improved by 3–10 times in calibration accuracy compared with the raw data inter‐calibration results. The model has been running for testing in the FY‐2B ground operational system and will be applied operationally to the follow‐up satellite system. With the adjustment of a few parameters, the principle of the model can be used for stray light analysis of other instruments on geostationary satellite platform.