Finite Element Analysis (FEA) is a widely used computational tool for evaluating Von Mises stress distributions in bones and joints, particularly in femoral Von Mises stress analysis. Despite its accuracy, its computation for large models requires substantial resources and takes considerable time, which limits its applicability in real-time simulations and large-scale studies. To overcome this limitation, we present FEA-GNN, a Graph Neural Network learning framework for femoral Von Mises stress analysis that accelerates FEA simulations by 3–4 orders of magnitude while retaining a very small relative error compared to ground-truth FEA solutions, indicating its prediction accuracy is extremely high and close to the actual value. Our method formulates the finite element mesh as a graph, enabling efficient learning of Von Mises stress relationships from precomputed FEA solutions. The substantial reduction in computation time makes it a promising alternative for biomechanical simulations. More specifically, a real-time simulation can be conducted with the FEA-GNN and a visualization tool has developed. This work demonstrates the potential of GNNs to revolutionize computational biomechanics through fast and reliable Von Mises stress predictions.