The response characteristics of CMOS pixels will be affected by factors such as production process, dust and damage, which in turn affects imaging quality. It is necessary to develop a method that can efficiently and quickly detect the location and area size of abnormal pixels in image processing. The response characteristics of CMOS pixel is analyzed in this paper, and a pixel classification method based on K-Mean algorithm according to the difference of pixel value is proposed. This method both effectively detect the abnormal pixels and classify them. Experimental results show that this method can effectively detect the types, position and size of abnormal pixel. Compared with traditional “3[σ]”method , this method run with faster detection speed in finding the abnormal pixel region and more accurate positioning.