Termination reflects desirable behavior property of active database systems. At present, termination analysis is based on triggering and activation graphs, the result of which is conservative. In this paper, “deactivation graph” is introduced to express the fact that one rule's action may falsify another rule's condition. In combination with triggering and activation graphs, a more generalized “relationship graph” is defined, based on which a new algorithm for termination analysis of active rule set is proposed, which improves the accuracy of termination analysis for active rule set.