The power grid dynamics in North America is online detected by the North American power grid frequency monitoring network, and the measured frequency signals by frequency disturbance recorders (FDRs) contain the white noise and spikes that affect the analysis to those signals. In this paper, an interference model is built to depict the FDR frequency signals. Since those contaminated and complicated FDR frequency signals cannot be denoised satisfactorily by a conventional median filter with fixed window size, it is necessary that an adaptive median filter is designed and employed to eliminate noise and spikes. Furthermore, the accurate event start time (EST) is very important to analyze the dynamic process of a power grid and to locate the hypocenter of both the load trip event and the generation trip event, and hence a simple and useful algorithm also is presented to estimate the EST. Two event cases from North American power grids are employed to validate our idea.