The semantic web goal is to share and integrate data across different domains and organizations. The knowledge representations of semantic data are made possible by ontology. As the usage of semantic web increases, construction of the semantic web ontologies is also increased. Moreover, due to the monolithic nature of the ontology various semantic web operations like query answering, data sharing, data matching, data reuse and data integration become more complicated as the size of ontology increases. Partitioning the ontology is the key solution to handle this scalability issue. In this work, we propose a revision and an enhancement of K-means clustering algorithm based on a new semantic similarity measure for partitioning given ontology into high quality modules. The results show that our approach produces meaningful clusters than the traditional algorithm of K-means.