Bias correction is the key technology of data assimilation for satellite radiance data. Currently, global GRAPES variational assimilation system adopts scan bias correction and airmass bias correction based on the consideration of Harris and Kelly (2001). But it did not cover the changes of deviation property (such as instruments or pollution problems). This paper proposed a scheme called online bias correction based on Harris and Kelly’s TOVS radiation error correction scheme and the concept of online bias correction which was put forward for NWP system to solve the problem of data drifting. The scheme considers the characteristics of GRAPES system and the statellite pretreatment system of National Satellite Meteorological Centre of CMA. The online bias correction is an adaptive bias correction system. It uses variational method to adjust the coefficients of predictors. To test the effect of the new scheme, twomonthlong assimilation experiments were carried out. The results show that the online bias correction scheme can optimize the degenerate bias equation automatically and quickly, and keep the effect of bias correction stable.