An automated pilling grading method was introduced. In this study, a pilling grade of fabric samples was evaluated visually by experts. A pilling grade was also predicted from a fabric height map using a convolutional neural network. The performance of this network model was evaluated by comparing it with the grade given by experts. The difference amongst the experts tended to be large around a grade of between 2 to 4. As a result, 60 % of the network model predictions were equivalent to the grade given by experts and 90 % of those were within the range of the maximum and minimum grade given by experts. Moreover, the correlation between the prediction and the average grade given by experts was as high as that between an expert and the average of the expert grade. The findings from this study show that the use of this network model could be an effective alternative to the conventional visual assessment.