Structured description of classes of patterns can be made using coincidence assembly of subpattern or topographical relations. The recognition cannot be done by a classical parsing algorithm and one is given which is both top-down and bottom-up; it starts its parsing from sure initial primitives. The analysis of such complex patterns requires the use of knowledge such as the possible context in which each subpattern appears. This paper emphasizes how such formal descriptions provide much information in order to speed up the analysis process. It also shows why the generative possibilities of a description must be separated from its descriptive power, in contradiction with what happens with string languages in the Chomsky hierarchy.