Due to focusing on template matching, conventional approaches bound their capability by complex and varied layout structures. This paper proposes a novel and efficient logical structure description algorithm, named structure description algorithm, to automatically extract logical structures from hypertext (Web) tables. Based on table field relationships, our approach starts from each data cell to search leftward and upward for its correlated headers. After that, rules for describing logical structure can be generated without defining the layout structure pattern in advance. In addition through the help of a table translation strategy, our method outputs a relational table which can be fed into a SQL database directly for information query and processing. Experimental results show that proposed method not only retains the logical structure in output relational table, but also outperforms two major methods on handling very complex Web tables.