With the Web of today being unstructured and semantically heterogeneous, keyword-based queries are likely to miss important results. Therefore, refining and expanding queries plays an important role today. This paper presents a novel approach for query expansion that applies dependency rules mined from a large Web crawl, combining several existing techniques for data extraction and mining. Additionally, the rules help binding keyword queries to the input fields of forms on Web portals, opening up the tremendous resources of the HiddenWeb. Preliminary, yet convincing experiments from a restricted application domain indicate the viability of our approach.