Abstract: The quality of numerical weather prediction depends closely on accuracies of initial condition provided by observation system. Satellite observations are very important source for data assimilating models, which is of good overcast, high resolution and stable, improving prediction compared to conventional data in many cases. A new generation polar-orbiting meteorological satellite of China, FY-3A is successfully launched on 27 May 2008, and FY-3B is also successfully launched on 5 November 2010. The two kinds of microwave vertical sounding sensors aboard are very similar in capability to ATOVS (the Advanced TIROS Operational Vertical Sounder) of NOAA series satellite. One of them is microwave temperature sounder (MWTS), and the other one is microwave humidity sounder (MWHS). They are used to sound the vertical distribution of the atmospheric temperature and humidity respectively. They provide very important observations for application in regional and global data assimilation system.Because of the observation instrument accuracy, observation operator approximation, assimilation model limitations, to the assimilating bias needs correction. The significant characteristics of bias effect are to cause systematic increment field, and the increment field could be obtained by OMB (observed minus background) statistic. This bias correction experiment of FY-3A satellite microwave data is developed on the basis of Harris and Kelley's bias correction experience method for TOVS radiation data, and with combination of improved WRF-3DVAR system. By analyzing the algorithmic method of radiative transfer in the spectral of microwave coverage, mapping function of model variables and brightness temperature of satellite microwave channel is established, making the fast radiative transfer model to be of quasi-linear expression, with considerable accuracy. Under the fast radiative transfer model and its corresponding tangent linear and adjoint model, a direct variational data assimilation system is established in the original assimilation framework using FY-3A microwave temperature sounder and microwave humidity sounder as input. In consideration of the spatial variations and the air mass dependence of satellite radiation data, the microwave data are processed with scan bias correction and air mass bias correction. And the microwave data of each channel basically has a fitting line along the leading diagonal after bias correction. Distribution of most the satellite observation and the brightness temperature derived by observation operator using background tends to be reasonable, and the bias is reduced a lot.With bias correction, FY-3A microwave data is directly assimilated in numerical weather prediction. The assessment of the forecast experiments for 4 typhoons shows that after assimilation the track forecasting ability is significantly improved, especially after 36 hours. And assimilation of FY-3A microwave data has reduced track forecast error by an average of 20%, while the assimilation of conventional data can reduce it by only 4%.