Independent Component Analysis (ICA) has became the most popular method to remove eye-blinking artifacts from electroencephalogram (EEG) recording. For long term EEG recording, ICA was commonly considered to costing a lot of computation time. Furthermore, with no ground truth, the discussion about the quality of ICA decomposition in a nonstationary environment was specious. In this study, we investigated the “signal” (P300 waveform) and the “noise” (averaged eye-blinking artifacts) on a cross-modal long-term EEG recording to evaluate the efficiency and effectiveness of different methods on ICA eye-blinking artifacts removal. As a result, it was found that, firstly, down sampling is an effective way to reduce the computation time in ICA. Appropriate down sampling ratio could speed up ICA computation 200 times and keep the decomposition performance stable, in which the computation time of ICA decomposition on a 2800 s EEG recording was less than 5 s. Secondly, dimension reduction by PCA was also a way to improve the efficiency and effectiveness of ICA. Finally, the comparison by cropping the dataset indicated that performing ICA on each run of the experiment separately would achieve a better result for eye-blinking artifacts removal than using all the EEG data input for ICA.