Abstract: The radiosonde observation of L band is a kind of conventional data, which plays an important role in weather forecast and numerical forecast. In recent years, with the progress of assimilation technology, the requirement of data precision also improves. It is found that more and more researches and operational forecasting centers are doing detailed classification and analysis for each type of instrument, even in each region and each season, according to the development of numerical forecast data assimilation technology. The method of humidity deviation correction takes the influence of observation pressure, temperature, solar altitude angle and other factors on the observation instrument into account and formulates a targeted deviation correction scheme to improve the data.Based on the analysis of humidity deviation and distribution characteristics of three kinds of L-band radiosonde instruments used in China, an effective correction method suitable for L-band radiosonde instruments in China has been developed and applied in GRAPES assimilation analysis. By improving the assimilation analysis and model prediction results, the observed humidity deviation of 3 kinds of instruments and results of continuous tests show as follows.Three main types of radiosonde instruments are used in China, among which Instrument 32 is most widely used, Instrument 31 and 33 are used at a dozen stations. The deviation of Instrument 32 is smaller than that of Instrument 31. The deviation of Instrument 33 is smaller when the humidity is greater than 60%, and greater than the others below the level of 400 hPa and the humidity is less than 60%.Compared with ECMWF reanalysis of humidity field, there is a dry phenomenon in L-band radiosonde humidity observation. Compared with the control test, the humidity deviation value of various deviation correction schemes is obviously reduced above the level of 500 hPa. The Vomel deviation correction scheme is used in GRAPES assimilation system, and the analytical deviation is reduced by 5%. After the humidity observation is revised, the forecast precipitation is closer to the actual situation, and the test score of forecast precipitation is improved significantly.Through the analysis and comparison of humidity observation deviations of several kinds of radiosonde instruments in China, the evaluation and understanding of the performance of these instruments are deepened, and the revised scheme suitable for radiosonde humidity deviation in China has been developed, which has achieved better application effect in the test. It lays the foundation for the better use of these data in practical applications, and makes an active attempt to better classifying sounding instruments.