The purpose of this study was to develop and test a training protocol for the perceptual evaluation of dysphonia. A group of 38 inexperienced listeners participated in a three-phase experiment: a pretest to evaluate their initial performance on categorization of dysphonic voices, a training phase, and a posttest to detect training-related changes in performance. In parallel, a different group of 14 listeners who were experts in voice assessment took a test that was identical to the posttest taken by the inexperienced subjects. The corpus used for the tests was made up of recordings of 142 voices of women reading aloud, with a sampling of voice qualities ranging from normal to severely degraded. The learners' performance on judgments of moderate and severe dysphonia improved between the pretest and the posttest. No improvement was observed for normal voices, whose initial detection was already good, nor for slight dysphonias, which appear to be the most difficult to learn. The improvements were still present on a delayed posttest taken a week later. Unexpectedly, the inexperienced listeners' initial performance was similar to that of the experts. After the training phase, their scores for severely deteriorated voices were even better than the experts'. In conclusion, our training protocol seems to be effective and could therefore be proposed to voice therapists. However, judging intermediate degrees of dysphonia remains fragile and therefore needs to be reinforced by repeated training. Key Words: Perceptual evaluation of dysphonia