For a lot of circulations of banknotes on market, most of the banknotes are unserviceable and the banks need to recycle or to reject these unusable banknotes. To recognize the banknotes fast and effectively, in this study, we develop a banknote recognition embedded system on Linux that is precise and efficient. The banknote recognition functions include banknote denomination, beveled edge of banknote and new/old of a banknote for New Taiwan Dollar. In the study, the banknotes are fed together into a Multi-slot Machine Banknote Sorter. Banknotes are scanned one by one from Contact Image Sensor (CIS) to ARM embedded system by the machine. Then, the banknote is recognized according to several features like banknote denomination, shapes of edges and brightness of the banknote image. Several experimental results are also included to show the average recognition speed is 360 banknotes per minute, success rate of beveled edge is over 97% and success rate of denomination is over 96% for undamaged banknotes.