Categorisation is a very common way we perceive our sound environment and, as so, was widely studied. Nevertheless, categorisation experiments are not so often used in applied studies. The goal of this paper is to show that it can be a very useful tool for reducing the number of stimuli without any loss of the generality of the context. That was checked in two cases : the first one dealt with door closing sounds. A large number of stimuli (35) was reduced to twelve thanks to a categorisation experiment; these twelve stimuli could then be used in a similarity rating experiment, enabling to know the perceptual space of such sounds. The second case involved sound computed from a simple model in which some structural variabilities were used. In that case, the perceptual spaces derived from the categorisation experiment and the similarity ratings one were very close to each other.