Financial fraud detection is becoming increasingly complex and vast, making detection difficult. This paper delves into the applicability of *Relational Graph Convolutional Networks (RGCN)* for fraud detection in models of heterogeneous graphs of financial transactions, where nodes signify entities like accounts and cards, and the edges denote their relations and correlations, for instance, the transactions. The RGCN framework learns patterns by aggregating neighborhood information based on relationships. Using a dataset with 4% fraudulent cases, the RGCN was able to attain an F1 score of 0.46 and a recall of 0.34 while effectively capturing localized fraud patterns. The study is thus able to demonstrate the possibility of using RGCNs in financial fraud detection, addressing class imbalance and scalability challenges. Future work will thus focus on enhanced performance and efficiency through optimized graph sampling and hyperparameter tuning.