Graph Neural Networks (GNNs) exhibit tremendous potential in addressing graph-related tasks such as node classification and link prediction. However, training GNNs on large-scale graphs poses computational and memory challenges for centralized training. Graph data is often distributed across multiple parties, and Federated Learning (FL) offers a collaborative approach to training a global model while preserving user privacy. The fusion of GNNs and FL, termed Federated Graph Learning (FGL), holds promise for privacy-preserving and scalable training of GNNs on large-scale graphs. Nonetheless, FGL faces performance limitations in scenarios characterized by non-Independent and Identically Distributed (non-IID) data, locally biased data, and the absence of efficient and secure communication protocols. The heterogeneity of non-IID data within local graphs can significantly degrade GNN performance, while locally biased data may lead to overfitting due to limited graph representation. Moreover, the inherent privacy challenges in FL impact the safeguarding of user data. This chapter presents FL and GNNs as separate entities, explaining the core challenges and recent advancements in their integration within FGL, focusing on data distribution and security efficiency. We specifically explore the applications of FL, GNNs, and FGL in fraud detection to illustrate how each framework converges in a real-world context.