Multi-category words are widely distributed in Chinese patent documents, and identification of them has been one of the difficulties in machine translation (MT). This paper proposes a rule-based method for identifying verb and preposition multi-category words in Chinese-English patent machine translation. Based on principles of boundary perception and according to syntactic and semantic information of multi-category words as well as context information, some reliable disambiguation rules are designed to help the MT system analyze proper categories of words, then proposes adverbial and predicate identification rules to determine and identify the words further. Related experiments and BLEU evaluations show that the method is efficient to recognize verbs and prepositions better, and is also helpful to improve final translation quality.