This paper is intended to present a novel rough set based approach to identifying base noun phrase (BaseNP). In this approach, we divide the whole task into two ordinal sub tasks: tagging and identifying. We regard BaseNP tagging as a decision + making process, which can be accomplished through rough set theory. What characterizes our tagging procedure is feature reduction and rule optimization. The focus of this paper lies in three aspects. First, we present a description of rough set-based rule learning mechanism and concerning algorithms. Next, we give a detailed account of the finite state transducer (FST) for BaseNP identification. Finally, we discussed the handling of instance collisions for improving system performance. Experimental procedures are described in detail and results indicate that rough set-based approach shows good prospects in natural language processing (NLP). At the end of the paper, we discussed the shortcomings of this approach and put forward suggestions as to its improvement