In order to investigate the time-varying non-linear characteristics of low frequency oscillations（LFOs）， a hybrid technique based on generalized mathematical morphological（MM） filter and Hilbert-Huang transform（HHT） is presented. First of all， the influences of different kinds of noises on HHT are analyzed. And then the generalized MM filter is built based on its principle and ability to filter the mixed noise. The data from wide area measurement system（WAMS） are utilized as the input of generalized MM filter， which can eliminate mixed noise rapidly and effectively. The filtered signals are analyzed by HHT， and then the instantaneous parameters of the dominant oscillation are calculated. Simulation results of the testing signal and practical signal from WAMS validate the presented hybrid estimation method.