The diagnosis of ADHD relies on psychiatrists’ knowledge and subjective experience. Many studies aimed to develop an objective method to assist diagnosis, but the performance of classification between ADHD and controls was not acceptable for clinical use. Here, we proposed a neural network model based on white matter information to classify ADHD and typically developing controls. Diffusion spectrum imaging and tract-based automatic analysis were used to measure properties of white matter. The neural network classification model was developed with high accuracy in the training and test data. It might be helpful to provide an objective way for diagnosis of ADHD.