Data collected during two Global Positioning System campaigns in 1994 and 1996 across Chile and western Argentina (22 stations), in the area where the 1960 May 22 Valdivia earthquake took place, shows ground motion velocities that cannot be fully explained by the elastic strain accumulation during the interseismic phase of an earthquake deformation cycle. We use dislocation models to reproduce the observed velocities, with a 3-D source in a medium with one elastic layer overlying a Maxwell viscoelastic half-space, and a planar rupture surface with uniform coseismic slip. The reason for avoiding a more detailed and elaborated model is that knowledge about the Valdivia earthquake source parameters and the area where the event took place is poorly constrained. We focus, therefore, on examining the first-order post-seismic deformation, and ignore finer details about the heterogeneity of the Earth. By means of a grid search inversion over more than a million different models, we derived the most likely values for some of the medium and source parameters involved in the deformation process, namely viscosity (η), thickness of the elastic layer ( D), average slip on the rupture surface and the seismic coupling coefficient (χ). According to our study, the optimum values are: and . A clear difference is seen between the surface deformation caused by silent-slip on the rupture surface and the one caused by post-seismic relaxation processes, two possibilities proposed to explain the anomalous velocities. We find that the deformation associated with the 1960 Valdivia event can still be observed after several decades and it is the most likely explanation for the velocity component that cannot be explained by plate convergence. Our model also predicts that this deformation will still be measurable for several more decades. Our model reproduces the first-order pattern of the measured GPS velocities, showing good agreement with recent finite-element studies, with the advantage of simplicity and short computation time, allowing the extensive search for the best-fitting model. [ABSTRACT FROM AUTHOR]