The brightness temperature bias of the medium wave channel of the variational assimilation geostationary interferometric infrared sounder (GIIRS) of FY-4 is required to meet the Gaussian distribution, so the bias correction of GIIRS data is necessary. Based on Harris B A and Kelly G ′s "off-line" method, a method for GIIRS bias correction based on the random forest is developed in this paper. In the specific implementation process, the cloud detection of GIIRS data is carried out based on the advanced geosynchronous radiation imager (AGRI) cloud products of FY-4. The experimental results show that the brightness temperature bias of GIIRS satisfies the assumption of Gaussian distribution after the bias correction. Compared with "off-line" method, random forest method has a better correction effect.