Sounding observation data are the most basic and important data source in weather forecast and numerical weather prediction. Vaisala Radiosonde 92/90 data account for a large proportion of this kind of observation. The quality of the humidity observation data directly affects the assimilation analysis and model forecast, so it is necessary to evaluate the quality of the data and correct the deviation. The humidity observation deviation of RS92/90 detector is found by comparing the humidity field of NCEP with EC reanalysis. Based on the deviation correction scheme of Yoneyama et al (2008) for the humidity observation of the instrument, the revised scheme, which is the most significant for the analysis and improvement and takes into account the influence of solar radiation and other factors, is selected from the comparison of several revised schemes. The scheme is applied to GRAPES-GFS test. The experimental results show that after the deviation of humidity observation for the RS92/90 sounding is corrected, the assimilation analysis and prediction have neutral effect. After the correction of prediction anomaly correlation in continuous experiments, there is an improvement in East Asia and the Tropics. The forecast score is slightly improved in these areas, so the correction scheme of humidity observation deviation of RS92/90 sounding instrument has its application value in practical operation.