Abstract: Monthly average temperatures from 1999 to 2006 at Tazhong weather station (39?0′N, 83?0′E) in the hinterland of the Taklimakan Dersert, and from 1961 to 2006 at 27 weather stations in the peripheral area of the desert, as well as the 2.5 　?.5 ?grid data of the NCEP/NCAR reanalysis surface monthly average temperature in the same period were used to reconstruct the time series of monthly average temperature at Tazhong from 1961-1998 by using the stepwise regression and empirical orthogonal function (EOF) analysis methods. The fitting errors, decadal mean temperatures and their trends were compared with one another. The results show that both the stepwise regression and EOF methods can be used to reconstruct series of monthly average temperature at Tazhong, but the former is superior to the latter in relatively smaller errors; the former's average absolute fitting error is 0.3℃ and the maximum absolute fitting error is 1.9℃. However, the NCEP/NCAR reanalysis surface monthly average temperature data can not be used due to its large positive systematic errors in winter season.

Key words: reconstructed series of monthly temperature, climate change, Taklimakan Desert