Despite the success of Graph Neural Networks (GNNs), they have not been sufficiently addressed from the perspective of transfer learning. Since graphs can attain a variety of structures or features in numerous domains, it is crucial to study the transferability between distinct graphs. In this paper, we explore the node-level transferability of GNNs using synthetic graphs generated by Stochastic Block Model. Our comprehensive experiments on a wide range of synthetic graphs with five GNN models reveal the characteristics of transferability of GNNs, including the influence of the graph structure and the feature information. We also examine the knowledge transfer from synthetic graphs to various real-world graphs.