total least squares-estimation of signal parameters via rotational invariance techniques） based method for the analysis of low frequency oscillations in power systems is presented. The method is one of the subspace-based high resolution techniques， which decomposes the signal into signal subspace and noise subspace from the data matrix that is directly made up of measured data. The method is capable of extracting crucial dynamic characteristics from a system of any size with high precision， and the information can be further used for designing damping controllers. The theoretical aspects of the TLS-ESPRIT analysis method are firstly revisited briefly in the paper. Then， to illustrate the proposed technique， three test cases are investigated， including analysis of a synthetic signal， a four-machine two-area power system and the New England 10-generator 39-bus system. Simulation results confirm the validity of the method and show that it has great potential as an alternative powerful tool for online and offline analysis of the low frequency oscillations in power systems.