Abstract: Ground-based microwave radiometer (MWR) can detect temperature and humidity profiles continuously and steadily, which compensate the shortcoming of the conventional sounding because of the long observation time interval. As a result, it is very helpful to explore the thermal process evolution of meso-scale synoptic system. At present, many types of ground-based MWR are developed at home and abroad. They are of different technical systems and their suitability for wide operational use is much concerned in scientific research institutions and management departments.The error of MWR product includes the contribution of both algorithm and hardware system, which is hard to distinguish. Therefore, to evaluate the observation performance of hardware system of the MWR, the brightness temperature of MWR is directly compared in this experiment. Using observations of 4-type radiometers and operational sounding data at the testbed of China Meteorological Adminatration from January 2016 to March 2018, and the simulated brightness temperature based on forward calculation from sounding data of MonoRTM as the reference, the accuracy of radiometers in different weather and seasons is compared and analyzed.Results show that the accuracy of brightness temperature of the domestic radiometer is similar to that of the imported radiometer. The observed brightness temperature of 4 radiometers are well related with simulated brightness temperature, and correlation coefficients basically are above 0.9, reaching a significant level of 0.001. Under clear sky conditions, the average of mean square root between the observed and simulated brightness temperature of four radiometers is 2.08-3.75 K. And the MWR-G shows the smallest error of brightness temperature, whose average deviation of each channel is 1.08 K, and the root mean square error is 2.08 K. The brightness temperature errors are minimum in winter and maximum in summer. Under precipitation conditions, the effectiveness of the brightness temperature observation of MWR is obviously reduced.Certainly, there are also some errors in sounding data itself. And it is difficult to completely avoid the drifting problem of sounding balloon, although a variety of ground-based remote sensing methods are used to assist the identification. It suggests to develop and apply calibration system with high accuracy and high stability, to ensure the accurate measurement of the radiometer. In addition, best observation mode of MWR during precipitation, and the material selection, replacement and maintenance of the radome need to be tested and verified, to expand the effective detection range of MWR.