|Abstract:||The Web has been rapidly ``deepened" by massive databases online: Recent surveys show that while the surface Web has linked billions of static HTML pages, a far more significant amount of information is ``hidden" in the deep Web, behind the query forms of searchable databases. With its myriad databases and hidden content, this deep Web is an important frontier for information search. In this paper, we develop a novel Web Form Crawler to collect the ``doors" of Web databases, i.e., query forms, to build a database for online databases in both efficient and comprehensive manners. Being object-focused, topic-neutral and coverage-comprehensive, such a crawler, while critical to searching and integrating online databases, has not been extensively studied. In particular, query forms, while many, when compared with the size of the Web, are sparsely scattered among pages, which brings new challenges for focused crawling: First, due to the topic-neutral nature of our crawling problem, we cannot rely on existing topic-focused crawling techniques. Second, existing focused crawling cannot achieve the comprehensiveness requirement because it is not able to be aware of the coverage of crawled content. As a new attempt, we propose a structure-driven crawling framework by observing structure locality of query forms-- That is, query forms are often close to root pages of Web sites and accessible by following navigational links. Exploring this structure locality, we substantiate the structure-driven crawling framework into a site-based Web Form Crawler by first collecting the site entrances, as the Site Finder, and then searching for query forms within the scope of each site, as the Form Finder. Analytical justification and empirical evaluation of the Web Form Crawler both show that: 1) our crawler can maintain stable harvest and coverage throughout the crawling, and 2) compared to page-based crawling, our best harvest rate is about 10 to 400 times better, depending on the page traversal schemes used.