Attention Deficit Hyperactivity Disorder (ADHD) is one of the common diseases of brain and has brought the growth of teenagers and even the adult indelible damage. It is very different to classify the ADHD symptoms and normal by the existing research. In this paper, the contributions are as two aspects: one is that the attributes of brain network of the resting state fMRI data have been calculated to discriminate three categories ADHD from the controls. And the average accuracies of various categories are 42.49% and 63.46% on the ADHD-200 datasets of NYU and KKI respectively, which is better than the average best imaging-based diagnostic performance of 35.19% and 61.90% achieved in the ADHD-200 global competition. The other one is that we put forward a new method named G-algorithm to construct the whole brain network, which based on certain rules. The same or even better classification results have been achieved by this method, which also verifies its feasibility and effectiveness.