During the development of a car, auralization allows listening to the expected sound of the prototype. This needs the computation of transfer functions. How detailed should these transfer functions be in order to provide a sound which is close to the real one? This paper addresses this question through an experimental process. A measured transfer function (between a force input to a chassis and a microphone in a truck cabin) was simplified by different ways (reduction of frequency resolution, averaging, resonance displacements…). Transfer functions thus obtained were used to filter an engine source signal. The resulting sounds were presented to listeners, whose task was to detect a difference between each sound and the reference one (i.e. the one obtained with the original transfer function). Results indicated that, in the middle and high frequency range, this difference is hardly detectable even when using strongly modified transfer functions. In the low frequency range, as the source signal has a strong harmonic structure, resonance frequencies can be easily detectable. Results were analysed using a loudness model (from Moore and Glasberg) and can give first ideas about how accurate a transfer function should be in order to be used for auralization.