Abstract: EOG(Electro-oculogram) can reflect the pattern of eye movement under different activities, therefore, reading activity recognition based on EOG has become a new research hot spot. To reduce the influence of blink EOG signal and improve the recognition accuracy ratio, a method based on ICA(Independent Component Analysis) was proposed, which is used to remove the blinkEOG signal. To start with, ICA was used to separate the blink signal from original multi-channel EOG signal. Then the algorithm not only identified blink signal channel automatically by calculating the kurtosis of different output channels, but also set all data of blink channel to zero. Finally, the processed channels were mapped back to reconstruct new multi-channel observation signals. In label environment, the average recognition ratio of the proposed algorithm reaches 95.5%. Experimental results reveal that performance obtains the relative increasing of 3.39%, 5.0% and 2.7% compared with original signals, band-pass filter and PCA, which verify the effectiveness of the proposed algorithm.