In this paper, we aim to propose an eye blink artifact rejection technique for single-channel electroencephalographic (EEG) devices. Independent component analysis (ICA) is usually employed for eye blink artifact rejection. However, ICA cannot be used for the artifact rejection when we recorded EEG signals by single-channel EEG devices, since this method requires multichannel signals for source estimation. On the other hand, single-channel EEG devices have been attracted attention since 2000 because of its usability for measurement and portability. In this paper, we propose positive semi-definite tensor factorization (PSDTF) with 2 step learning method as a new eye blink artifact rejection technique for single-channel EEG analysis. We investigate its validity using signal-to-noise ratio (SNR) between eye blink artifact signals estimated by our proposed method or ICA. Average value of SNR across subjects and trials is 16.13dB. For the value, the validity of our proposed method for eye blink artifact rejection of single-channel EEG signals was confirmed.