This paper aims at color grading of walnut kernels using a machine vision system. In this study, 159 kernels were divided into four color class of 'Extra Light', 'Light', 'Light Amber' and 'Amber' based on international standard. In image processing step, the Laplacian of Gaussian filter was used to remove the background of the image. Twelve color features were extracted from images including mean and standard deviation of red, green and blue as well as hue, saturation and intensity. The level of contribution of color features were determined using Average Squared Canonical Correlation values. The intensity mean had the highest contribution followed by saturation mean and intensity variance. Various neural networks architectures were developed with different number of inputs from one to 12 features based on the ranking model. Using the nine most important features as input for ANN, the highest grading accuracy of 95.8% obtained in the optimum structure of 9-20-4. The grading accuracy for walnut kernels in the classes of 'Extra Light', 'Light', 'Light Amber' and 'Amber' were 100%, 92.31%, 90.91% and 100%, respectively. The results show high potential of the machine vision combined with artificial neural network for color grading of walnut kernel.