The spread of fake news has become a worldwide problem, affecting public trust. Recent studies have reported that fake news and real news spread differently on social media. Thus, propagation-based detection methods, which construct graphs with users as nodes and news sharing chains as edges and simultaneously learn propagation patterns and users’ preferences using Graph Neural Networks (GNNs), have attracted much attention. However, for extracting users’ preferences from the graph, it is a challenge to learn the relationship between unconnected nodes. In this paper, we propose a method for fake news detection using Graph Transformer Network (GTN), which can learn efficient node representations while identifying useful connections between nodes in the original graph. The effectiveness of the proposed method is confirmed by comparison experiments using the real-world dataset composed of Twitter data.