This paper proposes the use of two-dimensional context-free grammars (2DCFGs) for layout analysis of PDF documents. In Japan, audio textbooks have been available for students with print disabilities in compulsory education. In order to create accessible textbooks including audio textbooks, it is necessary to obtain the information of structure and the reading order of documents of regular textbooks in PDF. It is not simple task because most PDF files only have the information how to print them out, and page-layouts of most textbooks are complex. By using 2DCFGs, we could obtain useful information of regular textbooks in PDF for the production of accessible textbooks.