Abstract: In order to improve the accuracy of variety identification of delinted cottonseeds, this paper proposed a nonlinear identification method based on Back Propagation (BP) neural network and investigated three varieties of delinted cottonseeds, namely Xinluzao 36, Zhongmian 50, and Huiyuan 710. Applying to image processing techniques, the color and shape characteristic parameters of delinted cottonseeds were selected, and then through carrying out univariate analysis of such characteristic parameters, 9 characteristic parameters that had significant difference were selected and involved in the network training, which improved the training speed. Through training and comparison, it was found that the training error was the smallest when the training target was 0.02 under condition that training times was 3 000, and the node number of hidden layer was 12. After testing the test set, it was found that the comprehensive test accuracy rate was 90%, which indicated that the method was feasible to improve the accuracy of variety identification of delinted cottonseeds. The research can provide a reference for variety identification of other granular seeds.