Graph pattern is the key body of a SPARQL query. Prevalent SPARQL query treatment delivers queries to OWL ontology model directly. To obtain inference results, the graph pattern are matched with inference ontology model which is generated by ontology inference engine. Since an inference model occupies much larger space than original model, and can not be reused as inference requirement varies, this method is not suitable to wide deployment in large scale. Alternatively, this paper proposes a novel method which sends rewritten graph patterns to original ontology model to acquire inference results. This method features in the reuse of original model between users, and avoidance of the heavy workload caused by generating and storing ontology inference model. The paper defines the ontology inference rules which affect the query resolving, and addresses a detailed process for rewriting graph pattern based on these rules. A prototype system is implemented to compare our proposal with current approaches. The experiment results show our method's advantages in aspects of completeness, soundness and effectiveness.