Electroencephalography (EEG) classification is the key point of brain-computer interface application. How to find effective feature is the major issues in EEG classification. Although several effective methods like support vector machines or neural networks have already been applied to EEG classification, but these methods need a large amount of prior knowledge to find the features of the data. Since the brain electrical signal appears to be more susceptible to noise interference and there are wide individual differences, so that effective features are difficult to been found. Meanwhile, it is difficult to improve the accuracy of the EEG classification, especially in the advanced cognitive process in the cigarette craving. In order to solve this problem, we use convolution neural networks (CNN) to classify EEG of cigarette craving patients under different status of cigarette craving. Compared with the traditional method, CNN does not need to manually extract features. It can directly train the original EEG data. More importantly, it can satisfy the demand which is to obtain the real-time feedback in the cigarette craving treatment process for classification results.