To better assimilate Advanced TIROS Operational Vertical Sounder (ATOVS) radiance data and provide more accurate initial fields for a numerical model, two bias correction schemes are employed to correct the ATOVS radiance data. The difference in the two schemes lies in the predictors use in air-mass bias correction. The predictors used in SCHEME 1 are all obtained from model first-guess, while those in SCHEME 2 are from model first-guess and radiance observations. The results from the two schemes show that after bias correction, the observation residual became smaller and closer to a Gaussian distribution. For both land and ocean data sets, the results obtained from SCHEME 1 are similar to those from SCHEME 2, which indicates that the predictors could be used in bias correction of ATOVS data.