A grape grading system based on computer vision was developed. The system consisted of 5 parts: the driving unit, conveying unit, holding unit, image grabbing-processing unit and grading control unit. Through a special chain, bunches of grapes were hung and conveyed continuously along a path. Two CCD cameras were installed to capture two images from two sides of each bunch simultaneously under external trigger motion. The color feature was extracted on RGB space, the size and shape features were calculated from the projection area and pixel accumulation curve. Then external inspection and grading of grapes were conducted. Twenty bunches of grapes were graded on the system for three times. Experiments showed that the accuracy was 90% for color grading and 88.3% for size-shape grading. The grapes were not damaged during grading.