Abstract: Aiming at discrete points collected from three geomorphic area, the influence of search points and directions on the interpolation accuracy of DEMs was investigated using Root Mean Square Error (RMSE) to select the parameters properly when constructing DEMs with radial basis functions. The results show that: for the plain and hill sites, Natural Cubic Splines Function (NCSF) and Thin Plate Splines Function (TPSF) are of higher accuracy with search points between 10 and 21 (the maximum being less than 18 for NCSF in plain area) and search directions have a little impact on the results. For the mountain site, multi quadric function (MQF) is the best choice with search points between 16 and 21 and orienting in four-directions 45° rotation. The research is helpful to optimizing the interpolation parameters and acquiring better output DEMs.