In listening tests involving different loudspeakers and aimed at assessing the sound quality of these sound reproducing systems, the level is generally adjusted to compensate for differences in sensitivity. The loudness sensation must be alike for each system under test. Because of the non-stationary nature of the musical signals used as test material in loudspeaker ratings, loudness assessment by using the current models (Zwicker, Moore...) remains slightly inaccurate. In practice, loudness is often equalized by ear by the experimenter. This study deals with the comparison of various test procedures. The first experiment was a paired comparison of loudspeakers where short-duration stimuli were presented to listeners for preference ratings. In the second experiment, the same listeners were allowed to switch, at any time, from one loudspeaker to another one so that the proposed stimuli were longer. In both experiments the loudness was equalized by the experimenter. However, under normal listening conditions, the listener is usually free to adjust by himself the reproduction level. At last, in a third experiment, the listeners had the opportunity, at any time, to not only switch from one system to another one, but also adjust the loudness of the stimuli.