In this paper, the problem of handwritten digit recognition has been addressed. However, the underlying language is Persian/Arabic, and the system with which this task is a capsule network (CapsNet) has recently emerged as a more advanced architecture than its ancestor, namely CNN (Convolutional Neural Network). The training of the architecture is performed using the Hoda dataset, which has been provided for Persian/Arabic handwritten digits. The output of the system clearly outperforms the results achieved by its ancestors, as well as other previously presented recognition algorithms.