(in English) In order to investigate decoding methods in silent speech Brain-Computer Interfaces, 19-ch electroencephalograms (EEGs) were recorded during covertly and overtly speaking Japanese vowels with three student volunteers. The EEG averaging was carried out time-locked to visual stimulus onset. To extract different encoding at each electrode, a similarity measure between each pair of electrodes was calculated, by which electrodes were divided into six groups. Finally, we investigated which components of the obtained event-related potentials (ERPs) has discriminability among the vowels. Consequently, the early components of the ERPs might include phonological and articulatory information.