The detection of event-related potentials (ERPs) through electroencephalogram (EEG) analysis is a well-established method for understanding brain functions during a cognitive process. To increase the signal-to-noise ratio (SNR) and stationarity of the data, ERPs are often filtered to a wideband frequency range, such as 0.05–30 Hz. Alternatively, a natural-filtering procedure can be performed through empirical mode decomposition (EMD), which yields intrinsic mode functions (IMFs) for each trial of the EEG data, followed by averaging over trials to generate the event-related modes. However, although the EMD-based filtering procedure has advantages such as a high SNR, suitable waveform shape, and high statistical power, one fundamental drawback of the procedure is that it requires the selection of an IMF (or a partial sum of a range of IMFs) to determine an ERP component effectively. Therefore, in this study, we propose an intrinsic ERP (iERP) method to overcome the drawbacks and retain the advantages of event-related mode analysis for investigating ERP components. The iERP method can reveal multiple ERP components at their characteristic time scales and suitably cluster statistical effects among modes by using a tailored definition of each mode’s neighbors. We validated the iERP method by using realistic EEG data sets acquired from a face perception task and visual working memory task. By using these two data sets, we demonstrated how to apply the iERP method to a cognitive task and incorporate existing cluster-based tests into iERP analysis. Moreover, iERP analysis revealed the statistical effects between (or among) experimental conditions more effectively than the conventional ERP method did.