Abstract: Statistical energy analysis （SEA） is a widely used method for analyzing the high-frequency dynamic response of mechanical structures. The reasonable division of subsystems is one of the critical basises for SEA. In this paper， an automatic identification method of SEA subsystem based on order-reduced modal energy density and hierarchical cluster analysis is developed. First， the structural modal energy densities in the high-frequency band are obtained through the discrete finite element model. Then， the main features of the modal energies are extracted through the proper orthogonal decomposition. The similarity of the modal energy density between different elements is analyzed by hierarchical cluster analysis. Finally， the number of statistical energy analysis subsystems and the corresponding structural elements are identified. The T-shaped plates， I-shaped plates， and engine combustion chamber are taken as simulation models to verify the effectiveness of the proposed method. Simulation results show that the coupling relationship between components， the number of subsystems， and corresponding elements can be automatically identified by the proposed method. Then， the SEA model can be established efficiently and accurately.