A method for dividing a character image into lines, comprising the following steps: segment-dividing step for, in term of pixels, dividing a character image into a plurality of character image segments arranged sided by side, each segment having a predetermined width; pixel distribution statistic step for obtaining the pixel distribution statistic in each image segment, namely the number of black pixels in each pixel-row of the segment, and obtaining the pixel distribution statistic in the whole image, namely the number of black pixels in each pixel-row of the whole image; segment block forming step for dividing the image segment into segment blocks according to the pixel distribution statistic of the image segments and the pixel distribution statistic of the whole image obtained in the pixel distribution statistic step; line images forming step for integrating the divided segment blocks into line images. According to the above method, the accuracy of line dividing of a character image, especially the accuracy of line dividing of character images having some noise, is improved, whereby the accuracy of character recognition is correspondingly improved.