This paper describes an approach to elimination of inference channels in the database based on rough set theory. The approach builds an incomplete decision table on all data in the database, then analyzes and discovers all the relations between non-sensitive and sensitive data with rough set theory. According to these relations, rules of inference control are generated and used to modify the data queried by generic users dynamically and most parsimoniously so as to eliminate inference channels. Experimental result shows that the approach is scalable and preserves security of inference control while improving availability of the data in the database.