Abstract: Aiming at the update of association rules with respect to the changes of the transaction database,in this paper,we discussed an approach for updating the former components of valuable association rules.First,we analyzed the dependency relationships and uncertainty among the components in association rules.Then,we constructed the Bayesian network (BN) model,called rule BN,to represent this uncertain knowledge.As well,we proposed an algorithm for RBNs approximate reasoning based on Gibbs sampling,so that the update of association rules can be fulfilled.Experimental results show that our proposed method for updating the antecedents of association rules is feasible efficient.