Abstract: Temporal features of annual temperature and precipitation at 28 stations in Gansu Province from 1960 to 2003 are analyzed using nonparametric Mann Kendall method and the accumulated departure approach. Spatial characteristics of these two variables in the 1960s and 1990s are detected using geostatistical method. It shows that the annual temperature has a significant increasing trend, while the precipitation has an insignificant decreasing trend at 0.1 significance level. Annual temperature experienced with low level from the 1960s to mid 1980s and with high level from 1986 to 2003, and abundant precipitation occurred in the 1970s and relative low in the 1990s at 0.05 significance level. Gaussian model is the best theoretical model for temperature semivariogram. 25% and 32% of total spatial variability in temperature is resulted from spatial autocorrelation in the 1960s and 1990s. Spherical model is the best model for precipitation semivariogram. The incidence of precipitation in the 1990s is smaller than that in the 1960s, and smaller than those of temperature in both 1960s and 1990s, indicating that the effects of local micro climate such as topography and slopes on precipitation is greater than those on temperature. 78% and 80% of total spatial variability in precipitation is attributed to spatial autocorrelation in the 1960s and 1990s, showing intensive spatial autocorrelation in precipitation in Gansu Province. Compared with the spatial variations from 1960s to 1990s, temperature increased in most of the regions, including the south of Longnan, Longdong, the east of Longzhong, Hexi Corridor plain, local area of Beishan middle mountains and the Qilian Mountains area. Precipitation tended to decrease in most of the regions, especially in Longnan area and Gannan Plateau area, while the situations in some local areas of Hexi Corridor plain and Beishan middle mountains area are on the contrary.