A machine vision-based method for grading the quality and type of commercial potatoes was proposed to solve the problems of lowering the commodity value of commercial potatoes due to their mixed sales， time-consuming and laborious manual sorting， and low efficiency of grading. A potato image acquisition device was built to collect videos of potato， intercepting images of potato at equal intervals. The images of potato collected were corrected， and then image processing methods were used to obtain binarized images of potato. Edge detection was conducted on the binarized image of potato based on the quality characteristics of potato. Potato contour coordinate points were extracted and a three-dimensional model of potato was constructed. The volume prediction model of commercial potatoes was constructed with linear regression analysis， and the quality prediction model was obtained according to the density formula to grade the quality of commercial potatoes. Eight physical parameters including the length， width， aspect ratio， area， perimeter， roundness， eccentricity and convexity of the smallest outer rectangle of the potato area in the image were extracted based on the characteristics of potato type. The applicability of the principal component analysis（PCA） was judged with the KMO test and the Bartlett's test. PCA was used to downsize the matrix of the physical parameters. A prediction model for grading the type of potato was established by combining with the logistic regression analysis method to grade and detect deformities in commercial potatoes. An experiment of grading quality was conducted on 40 samples of potato with different sizes. 50 samples of potato were randomly selected for grading the type of commercial potato. The results showed that the accuracy of grading with the volume prediction model was 95%， 100%， and 95%， respectively. The accuracy of grading the type of commercial potato with prediction model for grading the type of commercial potato was 92% and 100%， respectively. It is indicated that the machine vision-based method for grading commercial potato proposed can be used for online detection of grading the quality and type of commercial potatoes.