Artificial Neural Network (ANN) can be applied to recognice pattern, particularly at the stage of data classification. This study used a multilayer perceptron backpropagation ANN, an unsupervised learning algorithm, to recognize the pattern of uppercase handwriting on the answer sheet of multiple-choice exams. The application of this network involves mapping a set of input against a reference set of outputs. In this research, ANN was trained using 8000 handwritten uppercase characters (A, B, C, and D) consisting of 6000 training data characters (1500 characters for each letter) and 2000 testing data characters (500 characters for each letter). The result showed that for the most optimal performance, the architecture and network parameters were 10 neurons in hidden layer, learning rate of 0.1 and 3000 iteration times. The accuracies of the result using the optimal network architecture and parameters were 90.28% for training data and 87.35% for testing data.