Recently game addiction of young people has become a social issue. Therefore, many studies, mostly surveys, have been conducted to diagnose game addiction. In this paper, we suggest how to distinguish levels of addiction based on EEG. To this end, we first classify four groups by the degrees of addiction to internet games (High-risk group, Vigilance group, Normal group, Good-user group) using CSG (Comprehensive Scale for Assessing Game Behavior) and then measure their Event Related Potential(ERP) in the Go/NoGo Task. Specifically, we measure the signals of P300, N400 and N200 from the channels of the NoGo stimulus and Go stimulus. In addition, we extract distinct features from the discrete wavelet transform of the EEG signal and use these features to distinguish the degrees of addiction to internet games. The experiments in this study show that High-risk and Vigilance group exhibit lower Go-N200 amplitude of Fz channel than Normal and Good-user groups. In Go-P300 and NoGo-P300 of Fz channel, High-risk and Vigilance groups exhibit higher amplitude than Normal and Good-user group. In Go-N400 and NoGo-N400 of Pz channel, High-risk and Vigilance group exhibit lower amplitude than Normal and Good-user group. The test after the learning study of the extracted characteristics of each frequency band from the EEG signal showed 85% classification accuracy.