Compared with flat textual documents, XML documents are annotated with many meaningful tags, which give information retrieval systems a clearer understanding on queried documents. In addition to structured query languages, such as SQL, XQuery and XPath, keyword queries are widely used for XML retrieval because of their simplicity and ease of use. Although a single keyword and its query intention may be ambiguous, two or more keywords can clarify the query intention if possible occurring contexts and interrelationships are considered. This paper proposes the XNodeRelation algorithm to understand users’ keyword queries in XML retrieval. In contrast to existing approaches, the study infers users’ query intention by taking into account both schematic and statistical information of the XML data and considering the possible occurring contexts and interrelationships of query keywords. Extensive experiments verify the effectiveness of this algorithm.