Gsearch is a tool for searching tagged corpora. Queries are formulated in two stages. First, the user specifies a context-free grammar which is used to parse a given corpus and to convert its tags into a standard set. Second, a search expression uses the words, terminals and non-terminals provided by the corpus and grammar. Structured output may be visualized with Ratnaparkhi’s Viewtree program or Calder’s Thistle system. Gsearch currently works with BNC, Brown, SUSANNE, WSJ, Frankfurter Rundschau, and NEGRA.