High definition digital images of tobacco leaves were obtained, which were made on removing the background and segmentation processing. All the components of tobacco leaf color, such as red, green, blue and luminance, were calculated, as new indexes quantitatively described the color distribution state of tobacco leaf surface. The box-counting dimension variation curve of every color component was plotted. A hierarchical cluster analysis was performed on tobacco leaf color with fractal dimension as parameter. The results showed that there was obvious difference in the box-counting dimension variation curve of every color component of tobacco leaves from different growing areas. Cluster analysis with fractal dimension as parameter was very well consistent with tobacco grade experts. These results indicated color fractal dimension, as characteristic index of color classification, could more accurately describe the fractal distributed features of image color.