Abstract: Analysis of the system specifications and sounding characteristics of ground-based and satellite-based microwave radiometer is described.Based on the principle of atmospheric radiative transfer, millimeter-wave propagation model 1993(MPM93) and atmospheric absorption model, a forward atmospheric radiative transfer model is presented with acceptable error.The calculated brightness temperatures are better agreed with the measurements of microwave humidity sounder (MWHS) and ground-based atmospheric microwave profiler.Through the comparison of microwave radiometer in different sounding types,the paper constructs an improved neural network combining satellite-based and ground-based observations.For the data from January to December in 2008, the experimental results show that in the whole altitude(0~10 km), the improved retrieval model has high resolution with smaller root mean square error (RMS), and the retrieval water vapor profiles agree more with radiosonde datasets.