A novel image analysis method was proposed in the tobacco leaf grading system based on transmittance theory. Both kinds of images were captured under reflected and transmitted light. The test showed that three pairs of color features from reflectance and transmittance image had very low correlations. Six color features were applied to develop recognition model by using Fisher's linear discrimination method. The result shows that the combination of both kinds of images obtained a higher classification rate than either single image type did. It is applicable to introduce transmittance image into the tobacco leaf grading system.