There has been an increasing interest in n since it is spotlighted as the standard for data representation and exchange in the Web. The structural information for XML documents serves several important purposes. In spite of its importance, the schema is not mandatory for XML documents. Thus, much research to extract structural information for XML document has been conducted. In this paper, we present a technique for efficient extraction of concise and accurate DTD for XML documents. By restriction of DTD content model using the mixed content model of DTD and XML Schema as well as applying some heuristic rules proposed in this paper, we achieve the efficiency and conciseness. The result of an experiment with real life DTDs shows that our approach is superior to existing approaches.