Forgery is one of the critical problems affecting cash transactions. Forged banknotes are becoming serious threats hampering the smooth transactions in Ethiopia. Hence, the availability of such fake notes in the market needs the automation of money transaction system. The banking industries are unable to fully utilize self-serving devices including ATM’s intensively. Nevertheless, banks have not yet utilized a reliable recognition system to identify forged banknotes. This calls for the development of a better authenticity verification system. In this study, we have examined the color momentum, SIFT, GLCM, combination of SIFT, color, and GLCM, and convolutional neural network as a feature extraction technique and support vector machine, K- nearest neighbor classifier, and feed-forward artificial neural network as a classifier to design Ethiopian banknote recognition system. In order to minimize the effect of noisy data, we have employed an intensive image preprocessing tasks, like image histogram equalization and adaptive median filterbased image denoising. From the experimental results, we registered 99.4% recognition accuracy when using the CNN model as a feature extractor and FFANN as a classifier in classifying Ethiopian banknote denomination. Again, for fake currency recognition, CNN feature outperformed the other feature extraction techniques with an accuracy level of 96 %. We, therefore, recommend that a further investigation on the CNN model using advanced architecture like GoogLeNet and ResNet with the larger dataset to study the banknote classification and verification system.