Collocation is a powerful technique in filtering the correlated part of a signal from noise. It is now a day widely applied in many research field (e.g. in Geodesy to estimate the geoid or in Photogrammetry to reconstruct a DEM). In this paper a 3-D collocation filter is developed from the theoretical and numerical point of view. The main theoretical problem, i.e. to find a suitable catalogue of positive definite functions, is solved and a proper set of positive definite functions for a 3-D process is defined; futhermore a package of FORTRAN 77 programs is developed to implement the method. To test this stochastic filter, a simulated 3-D signal was computed; the numerical procedure has proved to be effective.