Finding simple non-recursive base noun phrase is an impoortant step for many natural language processing applications this paper presents a new corpus-based approach． In contrast to previous methods for Base NP identification We employ a decision tree to identify Base NP． And a self-learning mechanism is intgrated into our model． Penn Treebank is as the training standard． Experimental results show good performances using our method． The method can also be applied to processing of any other language．