Motion retargeting refers to the efficient transfer of motion across models with different proportions by decoupling motion data from specific character topologies. However, existing research mainly focuses on retargeting torso movements of the human body, with insufficient attention given to hand motion. Our work focuses on hand motion retargeting and proposes a novel retargeting algorithm based on graph attention residual perception. To enhance the model’s generalization capability, we conduct cross-domain training using both the Mixamo animation dataset and the InterHand2.6M real-world hand dataset. Our method employs an extended hand model and innovatively incorporates a Graph Attention Network (GAT), which effectively captures the biomechanical priors between finger joints while preserving fine-grained motion details at the frame level. Quantitative and qualitative experiments on public datasets demonstrate that our model achieves superior performance in the task of hand motion retargeting.