A white noise excitation based identification technique is presented for estimating low-frequency oscillation parameters. The identification process is more safe and economical by using this technique instead of the traditional disturbance method（e.g. artificial fault and adding/removing components of the power network）. First， with white noise excitation， impulse response function can be calculated according to the cross-covariance function of system input and output variables. Then， based on the estimated impulse response function and using‘eigensystem realization algorithm'， a multi-input and multi-output state space model of the power system can be built. Moreover， the eigenvalues and the mode shape of the system are also obtained. It is known that the produce of false modes is impossible to be avoided for any measurement-based method. However， all the true modes and false modes can be distinguished by the analysis of power spectral density of measured data. Third， a simulation for a four-machine power system shows the validity of the method for each step. Finally， through modal cross validation the identified state space modal is further proved to be accurate.