Recently it is common for users to acquire through the World Wide Web a variety of multimedia documents. As the number of Web documents is dramatically increasing, we need to develop a multimedia document retrieval system that can support both structure-based retrieval and content-based retrieval. In order to support structure-based retrieval, we design efficient index structures (i.e., keyword, structure, element and attribute) and implement those by using the o2store storage system. For the content-based retrieval, we implement high-dimensional index structure for color and shape feature that is based on X-tree. Finally, we do the performance evaluation of our multimedia document retrieval system in terms of system efficiency, such as retrieval time, insertion time and storage overhead, as well as system effectiveness, such as recall and precision measures.