The NCEP/NCAR analysis data is widely used not only in climate change simu lation and prediction research but also in cloud detection of remote sensing are a as initial data. An systematic intercomparison and analysis by Cressman interpolation is made between the NCEP Global Final Ana lysis (FNL) dataset and 753 station observed land surface temperature and 1.5 m above air temperature data over China in 2005 in terms of time, season and geo g raphical position. Through this comparison, some differences existed between the NCEP and the observed data can be identified.Comparison with the observed data , land surface temperature of the NCEP data is lower than the observed data in g eneral, while 1.5mabove air temperature is overestimated in the north of China and is generally underestimated in the south. In terms of season and geographic al position variation, the reliability of the NCEP dataset is better in summer t han in winter and is better in the southeast than in the northeast and northwest . From the point of diurnal variation (from 2:00 to 20:00 in local time) in land and air temperatures there is a tendency that the positive value areas (NCEP an alysis>observations) shrink and the negativevalue areas expand. The normalized RMSE shows that the practicability of air temperature NCEP reanalysis for Chi na is better than the land temperature NCEP reanalysis.