Abstract: In some specific areas, a single condition of semantic retrieval cannot have the ideal results. A retrieval algorithm based on semantic correlation between different ontologies is presented. The algorithm firstly builds the domain ontology, and then analyzes the existing instances to find out the semantic correlation among ontologies by clustering algorithm. Besides, the evaluation data of the user of the instance which is obtained by the survey is used as the sample, with which the deep belief network (DBN) is trained to obtain the weights of correlation between semantics of different ontologies. Finally, the relevancy between the retrieved models and the model in database is computed and the models with higher relevancy are used as the retrieval results. With the retrieval algorithm, the designer can get more satisfactory model in retrieval homepage, which greatly shortens the retrieval time.