The paper presents an interpolation method for surface air temperature (SAT) based on data fusion of multiple sources.It should check whether there is a significant correlation between primary and secondary variables firstly.Three multivariate geostatistical algorithums which includes collocated cokriging (CCK)，simple kriging with varying local means (SKlm) and kriging with an external drift (KED) were introduced to incorporating ancillary information into the spatial prediction of SAT.The method was illustrated using monthly mean temperature data from more than 720 meteorological stations in China in August 2008，and cross validation was performed to evaluate the performance of the map prediction quality.The results show that:Accounting for both land surface temperature (LST) from remote sensing and digital elevation model (DEM)，used as ancillary spatial information in three algorithms，outperforms accounting for only one ancillary data.Among all different methods，SKlm and KED incorporating LST and DEM have produced the best results，this is because:(1) LST is better to indicator the local trend of SAT.(2) DEM prefers to indicator the global trend of SAT.(3) Both SKlm and KED considering SAT with a non-tationary spatial distribution have better performance than others.