The problem of fuzzy ontology integration can be divided into two phases: the matching phase between ontologies and conflicting resolution on fuzzy values. This paper focuses on the problems of the matching phase, which currently has exhaustive and heuristic approaches. While the exhaustive method has many matching errors, most of the matched pairs between the two ontologies are detected. The heuristic approach uses the ontology nature to trim the non-homologous pairs which can decrease significantly the number of mismatching pairs, but often skips lots of homologous element pairs. To overcome the disadvantages of the two approaches, we proposed to use the constraint satisfaction problem (CSP) for modeling the ontology matching problem. In particular, we introduce constraints and suggest optimal function to minimize the matching errors. In the experiment, the mismatching pairs between ontologies are significantly reduced by applying CSP for refinement.