Abstract Evoked potentials (EPs) in response to stimuli are recorded from a human scalp contaminated with noise. To improve the signal-to-noise ratio, averaging methods have been widely used for the recorded data. However, when the waveforms of EP for each stimulus are not identical, the average waveform of the EP deteriorates. Variation of the EP waveform to each stimulus itself is important information for the EP. In this paper, a recording method for single somatosensory evoked potential (SEP) waveform is proposed, in which three kinds of band-pass filters were selectively used during three specific time sectors for each interstimulus interval. For the late section of the interval, an EEG waveform prediction method was applied to eliminate contaminated alpha rhythm components. By using the proposed method, we were successful in detecting the single SEP waveform.