As a sensory element, sound appears particularly well suited to treating children with autism spectrum disorder. But evaluations of the use of music therapy in this population are relative rare, particularly within a psychoanalytic theoretical framework. This study presents the results of a study, carried out from this point of view, that evaluated this treatment and the processes that characterize it. Using a population of thirty-six children, we compared the efficacy of music therapy in relation to simply listening to music on three different scales (CGI, CARS, ABC), after twenty-five sessions and eight months of care. Our results demonstrate that music therapy is more effective than listening to music on the basis of CGI scores. The two other scales (CARS, ABC) point out that children in both types of group made progress, but show no significant difference between them. Through a qualitative and psychoanalytic analysis, we describe more precisely the processes that characterize the autistic child’s encounter with sound. Our observations reveal a common therapeutic trajectory. Its first phase is one of bonding, and is characterized by a confusion of internal and external worlds. The second is one of detachment, as primary symbolization processes are launched. The third is one of re-bonding with the symbolizing function of the object, encouraging the development of reflexivity. Finally, we propose a broader model in the form of an evaluation grid for sound-based treatment.