Abstract: A recognition algorithm for character images is presented by using centroid angles:candidate comer points of character images are obtained by using Harris algorithm.After that,new corner points with greater precision are calculated by clustering candidate corner points.Based on these new comer points,we can calculate the centroid of the character image and all the vectors from the centroid to new corner points.The centroid angle is defined as an included angle between two adjacent vectors.The feature of a character is chosen as a vector which is composed of centroid angles in a clockwise (or anticlockwise) direction around the centroid.The centroid angles are invariant with respect to rotation,stretching and translation in geometry.And it is a better way to normalize images.The experiment result shows that the recognition rate for character images is improved.The effectiveness of the method proposed is verified.