Calligraphy is one of the most famous traditional art in China. The Calligraphy copying practice is the inevitable phase when learning Calligraphy. Calligraphy character has structure and stroke attributes, such as length of stroke and the position distribution of subpart, which can identify each certain character. In this paper, we propose a graph neural network-based algorithm which can measure the similarity between two Calligraphy characters according to structure and stoke. Experiment shows that the proposed method gives satisfied results with respect to the similarity measurement for the Calligraphy copying practice.