Recently, a lot of research concerning biological signal analysis has been proceeded in various field. Especially, research about electroencephalographic (EEG) signals has attracted attention since the signal contains a mixture of endogenous brain activities. However, the EEG signal is often mixed physiological artifacts such as blink, oculogyration, heart beat or muscle activities. Therefore, it is important to remove these artifacts from the EEG recordings when researchers attempt to analyze brain activities accurately. An independent component analysis (ICA) has been used for removing blink artifacts with over 90 percentage of the repeatability for original signals. However, a common major drawback to the ICA is that this method can only manage overdetermined mixtures, which entails using some EEG electrodes. There is no rejective method for single-channel EEG signals. Therefore, in this paper, we proposed an artifactitious rejection method using single-channel EEG recordings and 2 step non-negative matrix factorization (NMF). We acquired the multi-channel EEG recordings from 3 subjects who blink every 5 seconds according to metronomic sound. Furthermore, we applied the NMF using single-channel EEG (Fp1) recordings to reject blink artifacts, and the ICA using multi-channel EEG recordings to obtain clean Fp1 signals. High SNR between reconstructed signals by the proposed method and the ICA was showed. Moreover, over 99 percentage of the repeatability for original EEG signals was indicated by the proposed method.