In this paper we present a model and an adaptation architecture for context-aware multimodal documents. A compound virtual document describes the different ways in which multimodal information can be structured and presented. Physical features are associated to media instances, while properties describe the context. Concrete documents are instantiated from virtual documents by selecting and synchronizing proper media instances based on the user context: the situation, the environment, the device and the available communication resources. The relations between the context features and the media properties are described by a rule based system. Keywords Hypermedia design . Adaptation . Rule system