Abstract: Forward modeling of satellite atmospheric sounding instruments is the foundation of satellite data assimilation and inversion. Currently, more uncertainties on the accuracy of infrared water vapor channel are still exist compared with temperature detection channel. A method based on the group training of profiles which are classified by their water content of air column is used, to try to improve the forward modeling accuracy in the water vapor channel. Determination of threshold value is based on principles of basic balance profile numbers after group classification. Many threshold values are utilized to classify profiles, only results of threshold for 0.045 kg·m^-2 are detailed and analyzed, which lead to quite similar experiment results. TIGR 43 profile library is used as the training sample to get coefficients for fast radiative transfer computation and NESDIS 35 profile library is used to test the precision improvement as the independent sample.Different coefficients got from the group training are used to establish RTTOV forward model of FY-3 IRAS and the channel brightness temperature is calculated using the corresponding profile and coefficients. The forward modeling accuracy of the FY-3 IRAS brightness temperature is get by comparison with line-by-line results, which are considered as accurate and reliable. Research results show accuracy improvement of the brightness temperature after group training. Improvement shows better in the low water content profile case, which up to 0.17 K in the 0.045 kg·m^-2 threshold experiment. Further analysis is executed to find the cause for the improvement in the group training experiment. The layer water vapor predictor optical depth of channel 11-13 are calculated and comparison is done between the forward model with and without group training. Results show better consistency of water vapor absorption and channel weighting function distribution, besides, the absorption of water vapor line and continuum is adjusted more reasonable considering the weighting function height. These provide positive influence on the improvement of forward modeling results.It puts forward a method which can improve the fast radiative calculation accuracy of infrared water vapor channel, further work should be done in respect of bias caused by the precision of water vapor molecule absorption line parameters and the input water vapor profile self error. Furthermore, it merely considers the water content of air column, the potential influence caused by the shape of the profile is out of consideration, which may be the cause for the negative effect in channel 13 and research direction of future work.