Résumé : In this paper we present the methods implemented in the CLARCS (C++ Library for Automated Registration and Comparison of Surfaces) library. This library allows some basic and high level processing on free-form surfaces, represented as point sets or meshes. Three methods are the "building bricks" of CLARCS; they allow (i) the rigid/affine/non-linear registration of two point sets, (ii) the computation of the mid-sagittal plane of one point set, (iii) the computation of a mean point set from several point sets, and the variability around this mean. These methods are all based on a common methodological framework, in which the point sets/meshes are represented either as a Gaussian mixture model or as a draw of such a model. We propose some applications of the methods implemented in CLARCS on different sets of medical data.