This paper presents a diakoptic-based distributed state estimation (SE) algorithm suitable for large-scale power systems. In the proposed approach the large-scale power system is divided into a certain number of subsystems by removing tie line measurements. Based on diakoptic theory, the SE problem is then partitioned into a number of subproblems which are solved for each subsystem using local computational resources. In subsequent steps, the intermediate subsystem SE solutions are sent to a central computer for completing the state estimation process by taking the tie line measurements into consideration. Phasor measurement units are used to make each subproblem solvable and to coordinate the voltage angles of each subsystem SE solution. Test results on the IEEE 14-bus test bed and IEEE 118-bus test bed are provided.