This study is part of the European project CLOSED (Closing the Loop Of Sound Evaluation and Design, [http://closed.ircam.fr]). The CLOSED project aims at providing new tools to develop a methodology able to create and evaluate sound design products. Among these tools, a set of sound synthesis models based on physical parameters have been developed so as to encourage sound creation. These models comprise for instance solid contact models (impact, friction...) and liquid models (bubbles...). The goal of this master thesis is to contribute to the design of a perceptive interface for these sound synthesis models. It focuses on impact models and on material perception (more specifically on the four following classes: wood, metal, plastic and glass). The aim is then to perform a perceptive classification of synthesised impact sounds according to the material category in order to achieve a mapping between the physical space (the model parameters) and the perceptive space (the four material classes). The many physical dimensions of the synthesis model and thus the infinity of possible sounds represent a challenge for this approach. Actually, classical perceptive experiment methods are not adapted to deal with such a large sound corpus. A possible field of investigation is active learning techniques that may solve this problem by reducing the required amount of sounds to define the boundaries between the material classes. This report presents a study on a reduced parameter space, on which both classical perceptive method and active learning techniques can be carried out, so as to evaluate the latter method. Subsequently to this validation phase, a new experimental protocol driven by active learning procedures would allow achieving perceptive experiment on larger sound corpus.