With the development and prosperity of economy, the circulation of paper currency is growing rapidly. The banknote numbers are always collected for classifying the bankbill and identifying the counterfeit ones.After studing on the edge detection results of Sobel and Canny algorithms, we present a new banknote number segmentation method based on edge detection and projection, which can be devided into three steps-the location of the area on the left of the serial numbers, the location of the area that the serial number and the denomination number located, the identification and segmentation of the serial number and the denomination number. The front versus back, as well as inversion, are also studied in the dissertation.