Based on in-situ observed summertime air temperatures (OT) from 194 stations in China, NCEP/NCAR reanalysis 2-m temperatures (NT) and ERA-40 reanalysis 2-m temperatures (ET) on a monthly basis, a study is performed of the periods, variance, interannual and interdecadal variability of these temperature series by means of linear analysis, variance and wavelet treatments as well as EOF decomposition. Results suggest that (1) the values, anomalies, and trends are very close between the ET and the OT; the NT are lower compared with the OT and prior to 1965 the NT magnitudes and anomalies differ significantly from those of the OT and the ET, with higher similarity of the ET to the OT than that of the NT. Thus, in making periodic analysis of surface temperature, the ET are superior. Caution should be taken when the NT are applied, but their departures are somewhat credible after 1965.(2) The ET have similar linear trend and spatial mode to the OT, the trend of the NT is much greater compared with those of the OT and the ET, and the ET have higher ability to reproduce temperature trends in China on a long-term basis than the NT, which show uncertainties in long-range temperature variations in China.(3) The ET and the NT are capable of revealing the spatio-temporal patterns and variations of temperatures in China, but the ET are more correct, especially when the interdecadal variability is investigated of the temperatures in northern China north of 30°N. It is found that the ET and the NT have comparable ability to show the interannual variations.(4) The differences before and after 1976 in China are obvious in the three temperature datasets, with almost identical (not entirely coincident) trends and spatial distributions between the ET (the NT) and the OT. In addition, the NT before 1965 have a drawback similar to that of “PSFC”, a problem that should not be ignored.