In this paper, we deal with the problem of modelling and recognizing composite objects. Recognition involves two stages: first, recognition of the components via their visual features and second recognition of correct positions of the components with respect to the composite object geometric model. The geometric model we use allows for deformable composite objects, as well as indefinite number of components. Furthermore, we use a description logic in order to model the conceptual aspects of the part - whole relationship. The uncertainty involved in recognizing the various components, as well as the uncertainty involved in recognizing the correct positions is modelled through possibility theory. A maximal specificity criterion is introduced for merging components into composite objects.