Air temperature is an important parameter observed in metrological stations, and there are many ways to improve the precision of air temperature interpolation result. In this paper by using the air temperature data at 67 meteorological stations in Jiangsu Province in the year of 2003 and digital elevation model (DEM) data with spatial resolution of 30m×30m, four common interpolation methods, including Inverse Distance Weighting (IDW), Spline with tension (Spline), Ordinary Kriging (OK) and Co-Kriging (CK), were used to interpolate the monthly and yearly mean air temperature and the precision of those four methods was compared by using cross validation method. The results showed that OK has a much higher precision than IDW and Spline, indicating the method accounting for spatial self-correlation is more accurate than others. Four auxiliary variables, including latitude, longitude, distance from the coast and elevation, were selected for CK, and correlation analysis showed that the monthly mean air temperatures are best correlated with latitude, and the three other variables followed. As the four variables are correlated with each other, principal component analysis (PCA) was conducted in this paper. The first principal component mainly representing longitude and distance from the coast and the second one mainly representing latitude were utilized as the optimized auxiliary variables for Co-Kriging interpolation in most months except March whose input is only the second one, April and July whose inputs are the second and the fourth principal components. The results indicated that the precision of CK which makes good use of related auxiliary factors is slightly higher than that of OK; while it is obviously better than OK where there are fewer stations and is a potential ideal method for air temperature interpolation. The results of this paper also showed that distance from the coast is a critical factor to the spatial pattern of air temperature in Jiangsu, China, which should be an auxiliary variable for CK.