This paper deals with the use of system identification techniques for estimating low-order black-box state-space models formodal analysis. It presents a method to estimate state-space models for large power systems with many controllable devices from noisy measurements. To excite the system low-energy pulses generated by the controllable devices are used. The input signals are the controllable set-points of the devices, the output signals are the speed signals of some generators obtained from Phasor Measurement Units (PMU).