Abstract: Named-Entity Recognition（NER） is a fundamental task for natural language processing and information retrieval.Literatures are full of person，location and organization names，while complex named-entities as book names and movies names are seldom referred.The authors focus on the recognition of such complex named-entities in query logs of a search engine.The authors roughly segment the queries according to their Web context and use the Web data to train a complex named-entities classifier.The authors use three different classifiers，which show that the methods have fairly good performance.