The meaning of model and description is discussed in the framework of machine perception. An artificial intelligence perspective is adopted, viewing a model as a Knowledge Representation Language (KRL), and a description as a construct of a KRL. Generalised inference is a key feature, inextricably connected to all kinds of models. The latter are grouped into three basic schemes: relational, propositional and procedural, each of which motivated by the need to capture relevant aspects of perceptual tasks. The distinguishing features of the schemes are outlined. Profound links are emphasized between relational/structural and propositional/linguistic schemes, in analogy with mental representations in human beings. The need to further combine the schemes into unified, hybrid representation languages emerges as a trend in present and future research.