Abstract: Traditional techniques of automatic proofreading mainly adopt stochastic methods on character or lexical level, and cause some limitation. By discussing the design thought and implementation methods of automatic proofreading for Chinese texts, a proofreading strategy of combining multi-levels of morphology, syntax and semantics is proposed, and consequently the errors that cannot be checked before are detected. The whole framework of the system of automatic proofreading is described and the operable implementation methods are introduced in detail. Designed for the text of unrestricted domains, these methods provide new thoughts for future development of proofreading for Chinese text.