An adaptation technique of multimodal interfaces based on a cognitive modeling of users is presented in this paper. The adaptation process tries to select a combination of modalities that best fits the cognitive profile of each user. This profile contains indicators of cognitive abilities and styles. This paper focuses on output modalities and applies the adaptation technique on hypermedia documents. The system presented is implemented using the XML and ASP technologies. The paper describes the supervised cognitive training program called HAPPYneuron[TM], examines the cognitive styles, describes the two main components of the output profile, and illustrates the adaptation process and the document structure. (Author/AEF)