Protein domain-domain interaction pairs supply functional information about the interacting proteins; and finding interaction motif pairs in protein-protein interaction database can deeply disclose the essence of the protein interaction. Up to now, there is little research work on prediction of interaction motif pairs within domain-domain interaction pairs. In this paper, we propose a new method to predict domain interaction motif pairs. We start from collecting contact segment pairs in the PDB protein complexes, and then use the contact segment pairs as seeds to iteratively cluster the protein-protein interaction database with the help of functional domains, finally we generalize the similar segment pair clusters to produce motif pairs. Using our method, we find 528 motif pairs.