Abstract: In classical rough set theory，a target concept can usually be depicted via static granulation，which cannot well characterize dynamic cognitive process in problem solving.A target concept and a target decision were approximated using positive approximation and converse approximation，respectively，and they have been effectively applied in clustering algorithm and rule extracting.Based on dynamic granulation principle，the concept of bidirectional approximation is proposed，some important properties are obtained.They have been applied for extracting decision rules from decision table in rough set theory.