Using CAPTCHA is a simple but convenient way to ensure user data security. It’s widely used in user authentication and user interaction. In this paper, CAPTCHA images from several typical websites were used as the research objects. The paper shows the whole process on image binarization, de-noising, dilation, splitting characters. Gives out the CAPTCHA images recognition algorithm based on edit distance which defines the string similarity. Experiments show that the proposed algorithm is simple, fast, robust performance and has a high recognition accuracy rate.