Concerning the limitations of the Weighted Edge Pruning (WEP) method in accuracy and matching efficiency, a Deep Web entity matching method based on twice-merging was proposed by introducing the concepts of self-matching and merging. Firstly, attribute values of each object were extracted to regroup objects for gathering objects with the same attribute value together, therefore, all objects could be divided into blocks efficiently. Secondly, the matching values between objects within a same block were calculated for pruning, self-matching detection, merging explicit matching to generate preliminary clusters. Finally, based on these preliminary clusters, matching relationships were further discovered by using the message passing between objects within a cluster and objects' attribute similarity values, which triggered a new round of cluster merging and updating. Experimental results show that compared with the WEP method, the proposed method, by detecting self-matching to automatically distinguish matching relationships and take the proper matching method, gradually refines the merging process to improve the matching accuracy; simultaneously, by blocking and pruning to effectively reduce the matching space, its system efficiency is improved.