Aiming at the ultra-low frequency oscillation or low frequency phenomenon in power system, it is necessary to identify the negative or weak damping gensets which cause the oscillation accurately to formulate control measures ensuring the safe and stable operation of power grid. Therefore, this paper proposes an effective identification method based on a total least squares-estimation of signal parameters via rotational invariance techniques (TLS-ESPRIT) and damping torque analysis (DTA). In this method, firstly the dominant oscillation mode of each generator is extracted from measured signals (e.g. rotor speed, rotor angle, and active power) at its side by means of TLS-ESPRIT, and then corresponding damping torque coefficient can be obtained through the least square fitting method to determine whether the generator is the oscillation source of this oscillation mode. Finally, case studies are carried out on the 4-machine 2-area system. Simulation results indicate that the proposed method can accurately identify the native damping units.