In order to evaluate the performance of reanalysis surface temperature over Jiangxi Province, the gridded temperature data of the JRA55, ERAInterim, ERA5 and MERRA2 reanalysis models during 1980-2017 is topographically modified and interpolated to the station level by the bilinear interpolation method, to examine the applicability over observed temperature in terms of meanstate deviations (bias, root mean square error and correlation analysis), annual trend and extreme temperature events (the number of high/low temperature days). The results show that the lapse rate of air temperature calculated in the study has a reasonable range and seasonal cycle, and is suitable for the topographical correction for hourly data interpolation. Based on the topographically modified results, JRA55 has the lowest bias and root mean square error, followed by MERRA2, ERAInterim and ERA5. As for the tendency, JRA55, ERAInterim and ERA5 are consistent with observation in magnitude, while JRA55 has better capacity in capturing the spatial distribution of the warming. In the perspective of extreme temperature events, the four set of reanalysis data can reproduce the annual fluctuations, but only JRA55 shows closest magnitude to the observation. As a conclusion, the applicability is: JRA55>ERAInterim>ERA5>MERRA2.