We propose the first label memorization framework for node classification in GNNs and investigate the relationship between memorization and graph/node properties.