Objective: A support vector machine （SVM）-based grading system for kernel-free white grape bunches was designed to counter the problems of low accuracy and efficiency of manual grading of kernel-free white grape bunches. Methods: The images were preprocessed using Gaussian filtering, edge detection, contour detection and other pre-processing methods, and the SVM model was used to extract the contour, color and other feature parameters of the kernel-free white grape bunches and compared the recognition effects under different parameters based on the SVM model. Results: It was shown that the best parameters of the model were Best c=2.00, Best g=0.24, coef （）=0 and d=3. The grading accuracy of the kernelless white grape bunches reached 96%. Conclusion: Compared with the traditional manual grading method, the reliability and stability of the proposed grading system had obvious advantages and could meet the grading requirements of kernel-free white grape bunches in practical production.