The use of computers in educational assessment is a widely explored territory. Several studies have been performed to show the effectiveness of computer-assisted assessment (CAA) and it has been accepted in various education sectors. However, due to the lack of sufficient infrastructure and other issues, the paper-based examination is still being used for educational assessment in many countries including India. Existing CAA frameworks require the examination to be conducted on a digital platform. So, these do not apply to paper answer books. We propose a two-phase framework for automatic evaluation of handwritten answer books. The first phase converts the answers written on papers to a digital form using a neural network-based handwritten answer recognizer. The second phase evaluates the answer to generate a numerical score. For this evaluation, we use a model answer-based approach where various levels of similarities between the model answer and a student answer are computed. To assess the performance of the developed system, we apply it to the evaluation of class VII Social Science Geography answer books of an Indian school. The experimental result shows that the proposed approach is quite promising.