Grouping is often intended as a general-purpose early vision stage which gathers together image features of perceptual salience, usually having a well-definable structure. This work addresses the problem of generic part-based grouping and recognition from single two-dimensional edge images following a strategy that employs generic part models at all stages: the key underlying idea is to perform a purposive grouping of simple parts and these parts can be conveniently represented by generic deformable part models. This paper describes the proposed computational method, which is more extensively treated in Pilu (1996).