A method for recognizing abnormal image is disclosed. The invention utilizes level comparisons of adjacent image lines to determine if there is any abnormal image amid an image picture and whether the abnormal image belongs to shading or LBB. The method comprises the following steps. First of all, two adjacent image lines having level values P.sub.i and P.sub.i-1 are selected. Then an absolute value of the P.sub.i and the P.sub.i-1 is calculated. Next the absolute value is compared with a value X. When the absolute value is smaller than X, then the image lines are determined as normal. On the contrary, when the absolute value is larger than X, then at least one of the image lines is determined as abnormal. Moreover, another two image lines having level values P.sub.i+1 and P.sub.i-2 separately adjacent the image lines having level values P.sub.i and the P.sub.i-1 are selected. An absolute value of the P.sub.i+1 and the P.sub.i-2 is calculated and the absolute value of the P.sub.i+1 and the P.sub.i-2 is compared to the value X. When the absolute value of the P.sub.i+1 and the P.sub.i-2 is smaller than X, then the image lines having level values P.sub.i and P.sub.i-1 are determined as shading. However, when the absolute value of the P.sub.i+1 and the P.sub.i-2 is larger than X, then the image lines having level values P.sub.i, P.sub.i-1, P.sub.i+1 and P.sub.i-2 are determined as LBB.