At present, the cleaning of wrinkled black melon seeds is performed manually, which has extremely low efficiency. As to the problem, a method for identifying wrinkled black melon seeds with photometric stereo is proposed in this paper. First, the accuracy of image capturing system was calibrated by constructing a testing entity surface. Then, as there was a great color difference between the central and the periphery area of the black melon seeds, the two parts of surfaces were separately reconstructed. Finally, experiments were done to determine transverse surface area as the identification feature, and the threshold is nine. It was shown that more than 90% of the wrinkled melon seeds could successfully be identified while the error rate is below 5%. From the experiments, it can be concluded that the algorithm of photometric stereo can be used to identify the wrinkled black melon seeds effectively.