With the growing access to heterogeneous and independent data repositories, determining the semantic difference of two ontologies is critical in information retrieval, information integration and semantic web applications. In this paper, we propose an ontology comparison tool based on a novel senses refinement algorithm, which builds a senses set to accurately represent the semantics of the input ontology. The senses refinement algorithm automatically extracts senses from the electronic lexical database WordNet (locally installed or online), removes unnecessary senses based on the relationship among the entity classes of the ontology, and specifies relations and constraints of the concepts in the refined senses set. The senses refinement converts the measurement of ontology difference into simple set operations based on set theory, thus ensures the efficiency and accuracy of the ontology comparison. Our experimental studies show that the proposed senses refinement algorithm outperforms the naive senses set construction algorithm in terms of efficiency and accuracy.