This study focused on the estimation of cognitive information for unnaturalness that human feel. Estimation of unnaturalness from the human brain activity is one of key technologies for quality evaluation. Previous studies demonstrated that the amplitude of the oddball P3 component depend on the naturalness of visual stimuli. However the method requires multiple measurements for estimating clear ERP signal. This study, therefore, aims to estimate cognitive states about naturalness from single-trial EEG signal of the visual oddball task. This study adopted individual component analysis as pre-processing method, principal component analysis as feature extraction method, and, support vector machines, as classification method. This study estimated human cognitive state about unnaturalness from single-trial EEG.