A key problem in environmental monitoring is the spatial interpolation. The main current approach in spatial interpolation is geostatistical. Geostatistics is neither the only nor the best spatial interpolation method. Actually there is no "best" method, universally valid. Choosing a particular method implies to make assumptions. The understanding of initial assumption, of the methods used, and the correct interpretation of the interpolation results are key elements of the spatial interpolation process. A powerful alternative to geostatistics in spatial interpolation is the use of the soft computing methods. They offer the potential for a more flexible, less assumption dependent approach. Artificial Neural Networks are well suited for this kind of problems, due to their ability to handle non-linear, noisy, and inconsistent data. The present paper intends to prove the advantage of using Radial Basis Functions (RBF) instead of geostatistics in spatial interpolations, based on a detailed analyze and modeling of the SIC2004 (Spatial Interpolation Comparison) dataset.