The subject of this thesis is the experimental study of one aspect of sound perception called psychoacoustic roughness. The question of the causes of the percept is first raised. The strength of the sensation of roughness depends on a wide range of acoustical parameters. Existing theories, which may be divided into "spectral" and "temporal" approaches, diverge according to which parameters are considered relevant. These approaches are confronted in two experiments by manipulating the phase relations among components in synthetic sounds. The results lead to a rejection of the spectral approach and to a modification of the temporal approach, based on a computational model inspired by physiological mechanisms. The link between the sounds traditionally used in the laboratory and the roughness produced by complex sounds is considered in a third experiment on the roughness resulting from the addition of elementary tones. It appears that the roughness of complex sounds cannot be simply deduced from the sum of the roughnesses of their constituent sounds. A mechanism sensitive to the correlation between temporal envelopes has to be taken into account. A functional model of the effect is proposed. Finally, roughness is studied in a musical listening situation with instrumental sounds. A theory proposed by Helmholtz links the notions of consonance and dissonance in Western tonal music with the roughness of the chords usually employed. Two new experiments propose an extension of Helmholtz's theory to a nontonal context. The study of roughness perception illustrates the variety of sensory and cognitive mechanisms that need to be taken into account if one wants to understand what happens between the instant a sound reaches our ears and the moment we can "hear" it.