Long queries are widely used in current Web applications, such as literature searches, news searches, etc. However, since long queries are frequently expressed as natural language texts but not keywords, the current keywords-based search engines, like GOOGLE, perform worse with long queries than with short ones. This paper proposes a query substitution and search result refinement approach for long query Web searches. First, we retrieved several short queries related to a long query from the users’ query history. Then, we constructed the short query clusters and selected the most representative queries to substitute the original long query. However, since searching relevant short queries may ignore contexts and terms in the original long query and thus obtain diverse results and neighboring information, we compared the contexts from search results with the contexts from original long query and filtered non-relevant results. The experiments show that our approach achieves high precision for long query Web searches.