In the present work strategies to cope with outliers detection are defined both for datasets stored on 2D lattices and on 1D profiles. Assuming that (Hawkins D. M., 1980): "the outlier is an observation which deviates so much from other observations as to arouse suspicious that it was generated by different mechanism" we studied the problem of outlier detection in digital surface models, as first preprocessing and validation step. The methods proposed and the tools implemented have been applied to digital terrain models (DTMs), gridded geophysical data (gridded borehole depths, seismic velocities, amplitudes and phases, magnetic data, gravity data) but their use can be extended to data within different fields, as long as they represent surface models described by grid stored information. We decided to implement the software to blunders detection by adding apposite tools in GRASS (Geographical Resources Analysis Support System). The validation techniques are characterized by a common localization procedure: we examine the entire dataset by considering only a small subset at a time. Our basic hypothesis is that the values in the moving window (the mask) are observations affected by normal distributed white noise. An interpolating surface (a-priori model) is computed from the points surrounding the center of the moving mask (suspected blunder). The choice of the model determines the residual between the observation an the surface at the mask central point P0 and therefore the capability to detect the possible outlier. The surface model can be obtained in the following ways: polynomial interpolation, robust estimation by using the median, collocation (or kriging). For each of them we define an associated test in order to decide whether the point P0 is a blunder or not. The paper focus on the first approach.