Some comparative analysis of the seven stations sounding data during the South China Sea Monsoon Experiment (SCSMEX) in 1998 and NCAR analysis datasets (during 5-25 May 1998 and 5-25 June 1998 ) within the atmospheric boundary layer. The results show that: In the atmospheric boundary layer, NCAR analysis datasets have a big error. The error of the potential temperature and specific humidity of the andysis dataset decrease with elevation; the error of the specific humidity is bigger than the error of the potential temperature with the analysis datasets, basically, the square deviation of the potential temperature are less than 1 K at the seven stations, the greatest square deviation is not more than 2 K, while the square deviation of the specific humidity are greater than 1.0 g·kg-1, the greatest error are more than 2.0 g·kg-1at some stations; preliminarily, it can illuminate that the different underlying surfaces lead to a relatively large impact on the error of the NCAR analysis datasets, the potential temperature of the analysis datasets are not precise in inlands and islands than the values of the littorals, on the contrary, the specific humidity is precise in inlands and islands than the values of the littorals; in the vertical direction, the range of the potential temperature square deviation is 0.98-1.33 K, and the range of the specific humidity square deviation is 1.39-1.60 g·kg-1.