Many studies have shown that industrial products can exhibit a great variability in their vibratory and acoustical behaviour; but the influence of this physical variability on the perception of sound emitted by these products has not yet been investigated. The aim of this study was to evaluate that influence in the case of a computed system, including a force source linked to a plate through three springs. The physical variability could affect the force amplitude, the stiffness of each spring, the thickness and damping of the plate. Two listening tests have been conducted: in the first one, a L18 fractional designed experiment was used for the sounds synthesis; 20 listeners had to evaluate the similarity between each sound and a reference one (corresponding to the nominal state of the system). The most important factors were identified (force amplitude, plate thickness and the stiffness of one of the three springs); psychoacoustical criteria representing the perception were Zwicker loudness and Aures roughness. In the second test, 38 sounds (including the 18 used in the fractional design) were categorised by listeners according to their timbre similarity; a three-dimensional perceptual space was built from the answers. The plate thickness appeared to be more influent in that case and was linked to the first dimension of the perceptual space; it was related to a spectrum balance which could be correctly described from the specific loudness curves. Loudness and roughness still described the two other dimensions of the perceptual space. The differences in the answers to these two experiments can be either caused by the context (stimulus range) or by the task. A possible bias could have made the listeners judge the annoyance instead of the similarity during the first test.