The variational assimilation channel brightness temperature of the Infrared Atmospheric Sounder (IRAS) onboard FY-3B weather satellite requires its observation simulation deviation to satisfy Gaussian distribution. Because the deviations contained in satellite data processing and numerical forecast modes do not follow Gaussian distribution, it is necessary to correct them. First, the preliminary quality control such as cloud detection is performed for the data from IRAS. After the brightness temperatures in 20 channels of IRAS are counted, it is found that the scanning deviation is symmetry at nadir. Finally, both the scanning deviation and the air mass deviation are corrected. The result shows that after correction, the absolute deviation values in top channels 1, 10 and near ground channel 14 are increased. The deviations and standard deviations in water vapor channel 13 and ground channel 20 are increased. The mean brightness temperature observation deviation in other 15 channels are reduced from 1.04 K to 0.30 K. Their corresponding standard deviations are reduced from 2.28 K to 1.99 K. The deviation probability distribution is more Guassian.