Abstract: This paper shows the design and realization of a Chinese person name automatic recognition system based on the statistics on the large scale persons name bank and persons name corpus. It classifies the surnames by their priority and studied the statistical laws of the context of the first 300 surnames as single word. It also considers the process of confirming true surnames and recognizing person names as a process of decomposing the surname word set of Chinese sentences. At the same time, it proposes multilevel threshold, that is, the threshold of the right bound of person names and recognition both are functions of surnames. In the end it evaluates the performance of the system on a real data set, the recall and precision are 80.62% and 89.27% respectively.