The industrial field of door closing sounds consists in translating drivers' expectations into specifications for the mechanical components of the door. A client who manipulates a door in a showroom and perceives from the door closing sound that it is indeed closed also induces complex evocations of quality and solidity to which the manufacturer is particularly attentive. The translation of these complex evocations into technical rules is not an immediate process. A first step consists in characterising what is perceived thanks to criteria extracted from the acoustical signal, and this is then completed by a characterisation of the organic sources and acoustic transfers in order to establish the desired technical specifications. This thesis has endeavoured to study the first step of the process, one that raises fundamental questions on the impact sound perception: this has consisted in understanding what we perceive from an action-related impact sound, and to extract the underlying criteria from the acoustic signal. In Situ experimentation is first carried out in order to observe the qualitative characteristics of the uses and the impressions derived from a natural situation, thus identifying the pertinent environmental factors. This phase is completed with a quantitative study of the influence of environmental factors, such as other door sounds, motion perception and a priori image of the vehicle. After observing the links between In Situ and laboratory perception, door closing sounds are finely decomposed into perceptual properties: analytical properties (which are obtained thanks to sensory analysis), natural properties (linked to perception of sources and events) and evocations. The acoustic characterisation of the sound is then processed by means of an analysis-synthesis model which aims, not at reproducing the exact replica of the door closing sounds, but to synthesize sounds that preserve perceptual properties with a reduced number of signal parameters. The model consists in decomposing the sound in several independent impact sources, each impact being modelled by a set of gains and damping factors in frequency bands. The model is specifically calibrated to reproduce the analytical properties that are more directly linked to the acoustic signal. The controlled sounds are then used to observe the effects of acoustic parameters on the perceptual properties, and to propose underlying acoustic criteria.