In current social tagging system, users can freely add tags for the uploaded resources, which caused a problem that many tags could not describe the resource properly and even have some spelling errors. This problem may bring unnecessary troubles for other users who want to search this kind of resource. In this paper, a tag recommendation system based on collaborative filtering and text similarity is presented to solve the problem mentioned above. This system can automatically recommend some relevant tags for the new uploaded resources and thus the users can freely select tags from the system. Experimental results show that the recommended tags can effectively represent the contents of the webpages marked. Compared with the existing tag recommended methods, this method not only improves the accuracy of tags recommended, but also facilitates the webpage sharing and retrieval.