Abstract: Ground-based microwave radiometer (MWR) detects atmospheric temperature and humidity by receiving atmospheric microwave radiation, which can conduct 24-hour unattended, high-resolution observation. It can detect short-time variation of atmospheric elements. MWR is an important supplement to routine sounding. However, it has different observation accuracy at different times, seasons and weather conditions. Observation accuracy and influencing factors analyses are essential in scientific experiments and operation processes.In order to minimize error effects of radiosonde migration, results from three types of remote sensing devices under two weather conditions are compared. By analyzing differences of temperature and relative humidity between sounding and three MWRs at home and abroad of different technology systems in no-cloud and cloud samples, performances of these MWRs are evaluated.In the aspect of temperature, the correlation coefficient for MWRs and sounding is above 0.98. In no-cloud condition, errors of MWR-G and MWR-A are less than ±1℃ (the former is negative and the latter is positive). MWR-A has -1.8℃ deviation. Root mean square errors (RMSEs) of three types of MWRs increase with height. RMSEs of MWR-G, MWR-A and MWR-C are 2.2℃, 3.0℃ and 3.8℃. In cloud condition, vertical distributions of temperature error of three microwave radiometers have no significant changes in comparison with no-cloud condition. The RMSEs of three microwave radiometers in cloud condition are 0.5℃ higher than those in no-cloud condition. Microwave radiometer can identify the near surface radiation inversion layer accurately. But it's hard to identify the high-altitude inversion layer. In the aspect of relative humidity, the error of cloud samples is higher than the error of no-cloud samples, and the error of middle-high cloud samples is higher than that of low cloud samples. In no-cloud condition, RMSEs of MWR-A and MWR-C are 15% and 18%, less than the RMSE of MWR-G. In cloud condition, RMSEs of MWR-G, MWR-A and MWR-C (about 26%) are larger than those in no-cloud condition. The existence of cloud has a major influence on the detection of microwave radiometer relative humidity:No matter which height level, errors of low-middle cloud samples are bigger than those of no-cloud samples. Errors of high cloud samples are bigger than those of no-cloud samples, and the amplification of RMSE is about 10%-20%. This comparison analysis will provide some reference basis for the further improvement of the accuracy of MWR atmospheric profiles and the scientific research, promotion and operation processes of MWRs.