The accuracy of synchronous machine parameters is extremely important to power system operation and control. Based on matrix pencil algorithm， a new method is proposed to identify synchronous machine parameters， with the basic principles and processes of identification provided. The method can structure two Hankel matrices from collected three-phase short-circuit current. The noise effect is restrained based on singular value decomposition（SVD） and low rank approximation. The computational costs are reduced. DC component and fundamental component are extracted as well using specific relationship between matrices， consequently accurate identification of synchronous machine parameters can be realized. The simulation results show that this method has higher precision， computational efficiency and stronger anti-noise capacity.