Abstract: Examining the external quality of tobacco leaves is now mostly relied on the sense of human being. A computer vision system of tobacco leaves grading was introduced in this paper. This system has some subsystems. The demarcation system can control and set the phototonus of CCD camera. The learning system is used to extract and select eigenvalue, then to constitute characteristic vector. It can also remove the freak samples from the standard samples. The neural network is used to learn and grade the tobacco leaves in several areas. And the results of grading indicated that the examination precision of computer vision is all above 80% coincided with grading of human being, and about a half of products from different areas are more than 90%. So this system is highly practical in tobacco leaves garding.