Current semi-structured keyword search and natural language query processing systems use ad hoc approaches to take advantage of structural information. Although intuitive, they are ultimately ad hoc. We have developed the concept of coherency ranking based on the statistical properties of data to rank the keyword search results. We demonstrate how the coherency ranking works for two real-world XML databases and show its advantages over the previously proposed XML keyword search methods. Categories and Subject Descriptors H.2.4 [Database Management]: Query Processing General Terms Algorithms, Design, Language