Abstract : Loudness, the subjective intensity of a sound is a paramount psychoacoustic parameter in the evaluation of noisiness and in studies on sound quality. There are currently models which estimate correctly the loudness of stationary sounds, such as those of Zwicker (1958) and Moore and coll. (1997). Other models (Zwicker and Fastl (1999) and Glasberg and Moore (2002)) exist to calculate loudness as a function to time but present some limits in giving the total loudness of nonstationary sounds and in particular of impulsive sounds. The aim of this doctoral work is to study the loudness of impulsive sounds in order to propose an estimator of it to avoid running psychoacoustic tests, wich are long and expensive, but are the only mean today for determining precisely the loudness of such sounds. The originality of this work is to be focussed on impulsive sounds having a fast attack, not stage and a decrease of exponential form. The choice of these sounds is based on the observation of natural sounds which are commonly found in our sound environment. A first algorithm (ESI) was proposed to calculate loudness from physical characteristics of the signal such as energy and decay time. We determined the equal-loudness contours, the loudness functions and the temporal summation of impulsive sounds not only to determine the relations between the loudness and the physical parameters of the signal (frequency, level and duration), but also to test the models and to determine their applicability. The results of our research being partly dedicated to the industrial field and metrology, we are mainly interested in the loudness level, expressed in phons. Indeed, this unit is more practical because it is closer to the sound pressure level, expressed in dB SPL or dB(A), used more frequently. Thus, we determined the method of measurement of the loudness level best suited to our types of sounds. Thus, the method of adjustment was chosen among four other methods, for being reliable, stable and quick. All these data contributed to determine applicability and limits of the already existing loudness models and our new estimator (ESI). These data also contributed to propose a new index (Nimp), calculated from the model of Zwicker and Fastl (1999), to give a first approximation of the loudness level of impulsive sounds.