【Objective】 To design a matched online grading system based on the structure and working behavior of the apple harvesting robot，and meet the needs of grading apples in real time. 【Method】The pregrading principle was proposed to eliminate apples with diameters below standard which could improve the grading efficiency. Apple weight was measured by a force sensor and the weight grade was determinated. Apple size and rot area were detected using the machine vision technology. The image processing algorithm and interface control program were developed using Matlab and VS2008. The distributed control network was constructed based on CAN bus. Comprehensive grading tests on apples were performed. 【Result】The determination coefficient of apple actual diameter and detected diameter was 0.990 3, the determination coefficient of the actual weight and test weight was 0.999 6, the determination coefficient of actual rotting area and detected rotting area was 0.985 5. The success rate of comprehensive grading reached 89.71%, and the average grading time per apple was 2.89 s during continuous grading.【Conclusion】The apple grading system is stable, easy to expand, highly efficient and accurate, and can meet the real-time grading needs of the apple harvesting robot.