In order to examine the perceptual dimensions used by listeners to differentiate sounds, dissimilarity judgments are analyzed with a multidimensional scaling technique. The paired comparison method is often recommended as the standard method for collecting judgments of dissimilarity between audio stimuli. An alternative method is presented here. On each trial, listeners are asked to select which stimulus among three stimuli is the most similar to a reference stimulus. The method was tested wuth simulations and an actual listening test. Across trials, every stimulus was used in turn as the reference stimulus. The analysis of simulated data indicated that, the number of stimuli did not influence the estimation of dissimilarity, and that 20 simulated listeners were sufficient for recovering continuous dissimilarity values. The listening test conducted to evaluate the musical restitution by 12 loudspeakers led to two perceptual dimensions similar to those obtained in a paired comparison experiment.