Based on the observation data from 8 field stations in the semi-arid region of China and North America during 2003 to 2008,the land surface flux data in NCEP reanalysis dataset was systematically assessed and corrected with statistical methods like linear regression,curve fitting and dual regression fitting. Results show that the fitting method is not capable enough in terms of daily average value, except for soil wetness and sensible/latent heat fluxes at some individual sites. In contrast,improvements at some degree are identified in terms of monthly mean value in all the three schemes. A long-term land surface flux sequence over semi-arid regions of China and North America is established by combining the advantages of both linear regression and curve fitting methods,which can not only reflect the linear trend of the original reanalysis data but retain the diurnal and seasonal variations to a great extent at the same time. The 30 a land surface flux dataset provides a series of useful information related to the regional climatic change closely. In addition,the reconstructed long term surface dataset is beneficial to better understanding the land-atmosphere interaction and numerical model validation in this region as well.