Based on hourly observations from more than 2400 national weather stations in 2010-2015, this paper evaluates and compares the surface soil temperatures of EAR70, CLDAS and ERAInterim. 〖JP2〗The main conclusions are: CLDAS surface soil temperature had the highest accuracy in space (ME was -0.5 ℃, 〖JP〗RMSE was 3.0 ℃, R was 0.96). EAR70 benefited from the initial land surface field with CLDAS’s highprecision ME improved. The accuracy of reanalysis surface soil temperature was significantly reduced at around 06:00 and in summer and autumn. Reanalysis surface soil temperature exhibited a cold deviation during the relatively highvalue period. The reason was that the simulated soil temperature value rose slowly, and the corresponding parameterization scheme needed to be modified. There was a cold deviation in the Northeast region where snow covered the ground in winter, which may be related to the snow cover. The land surface parameterization scheme needed to be improved. In the complex terrain of the QinghaiTibet Plateau, CLDAS, which integrated ground observations, performed a good precision in atmospheric forcing data and thus improved soil simulation. The resolution of ERAInterim was relatively coarse and not suitable for the QinghaiTibet Plateau or coastal areas. Benefitted from CLDAS’s soil states in good quality, the accuracy of EAR70 had been improved in the QinghaiTibet Plateau. The high accuracy of the soil state’s initial field decreased significantly following the time increased after observation entering the assimilation system. Therefore, the realtime assimilation method used in CLDAS can improve the accuracy of soil temperature signals effectively.