This paper reports the results of a query expansion experiment making use of semantic data produced by a program that automatically performs the distributional analysis of a large corpus. This method allows us to take into account relations that go beyond classic lexical relations. We look at these results from both a global and a local perspective, showing that such broader and corpus-based semantic relations may be useful, provided that the expansion is controlled by a filtering technique based on the texts of the database (distributional feedback), and by an analysis of the query linguistic characteristics.