Along with the widely used Internet, the research of multimedia data management becomes more and more important. The traditional information retrieval (IR) system that manages the alphanumeric data is changing to the multimedia information retrieval (MIR) system whose purpose is to storage and retrieval multimedia data. As the kernel of the MIR system, multimedia document database must provide a normal and strict data model to present the overall and detailed content information of multimedia data. This kind of presentation does not only contain the content based on the structured features (e.g. spatial relations) but also on the semantic features (e.g. the information "shake hands"). But most of current data models of multimedia document database have some shortage in providing the description of this information in different layers and different granularities. In this paper, we locate the content information of the multimedia documents into several proper layers and define MDU (multimedia document basic unit) to collect this information in different granularities. And then, we discuss the relations among MDUs. Combining the definitions above, we can formalize the presentation of the multimedia documents and multimedia document database. An example is provided to explain how to use this presentation method, and we also compare this method with related works. And, we explain how to query multimedia data on our model.