This paper proposes a collaborative approach, which combines two processes: the extraction of approximate information from a rough relational database and its generalization into symbolical rules. A rough relational database model is basically a standard relational database model extended with some of the essential features of the rough set theory. The rough approach to relational databases allows the user to represent imprecision in querying, which gives a greater flexibility to the querying and also, improves the representational power of a relational database. The paper describes the prototype system ROUGH-ID3, which implements a hybrid knowledge extraction approach by integrating a set of rough database operators with the symbolic system ID3.