Deserts cover a vast area of the world’s land surface, but the study of desert climate has been impeded by the lack of ground meteorological observations. In recent years, the climate reanalysis products provide an important data source to investigate climate change in observation-limited areas. However, their accuracy in desert regions has been poorly investigated. Here, we evaluated the performance of the latest ERA5 reanalysis datasets for the climatic conditions over the deserts of northern China (DNC), including temperature and precipitation variations, climate extremes, and detection skills of daily precipitation. The results show that ERA5 well captures the observed pattern of annual and seasonal temperatures, as well as the warming trend during the past four decades in DNC, compared to the ground observations. However, both annual and seasonal precipitations are greatly overestimated over DNC, and large uncertainties exist in precipitation trends. In addition, the variability of interannual precipitation, precipitation intensity, maximum 1-day precipitation, and the number of continuous dry days are underestimated by ERA5. The bias of precipitation estimates may be traced to the overestimation of rainfall occurrence in ERA5. Comparing to other reanalysis (such as MERRA2, NCEP/NCAR, and NOAA-20C), the ERA5 outperforms them for precipitation and precipitation extremes over DNC, although its performance is not as good as gridded gauge-based precipitation datasets (such as CPC). Overall, this study provides references for the use of ERA5 over DNC and facilitates the applications of ERA5 over other observation-limited desert regions.