The recognition of handwritten documents, which aims at transforming writtenrntext into machine encoded text, is considered as one of the most challengingrnproblems in the area of pattern recognition and an open research area. Especiallyrnancient manuscripts, like Ethiopic Geez scripts, are different from thernmodern documents in various ways such as writing style, morphological structure,rnwriting materials and so on. This brings the necessity to make researchrnworks on characetr recogntion of those scripts. Geez is one of the ancient languagesrnwhich has been used as a liturgical language in Ethiopia. Manuscriptsrnwritten using this language contains many unexplored content which is thernbase of the current Ethiopic scripts; however, only few researches have beenrndone on these valuable documents.rnA number of algorithms have been proposed for handwritten character recognitionrnsuch as support vector machine, hidden Markov model, and neural network.rnIn this research the design and implementation of character recognitionrnsystem for ancient Ethiopic manuscript using deep neural network is presented.rnDeep learning, is employed and trained using a Restricted BoltzmanrnMachine (RBM), a greedy layer-wise unsupervised training strategy.rnThe complete system employs image acquisition, preprocessing, character segmentation,rnand classification and recognition. Efficient and effective algorithmsrnwere selected and implemented in each step. A dataset was also prepared torntrain and test the system, which consists of 24 base characters of Geez alphabetrnwith 100 frequencies. Overall, a recognition accuracy of 93.75 percent wasrnobtained using 3 hidden layers with 300 neurons. Analysis of results obtainedrnirnfrom each step of the recognition process shows that the system can be extendedrnand fine-tuned for practical application.rnKey words: Ancient Ethiopic Manuscript, Handwritten Recognition, Preprocessing,rnsegmentation, Deep Neural Network, Restricted Boltzmann Machine.