Background: A dataset[1,2] of electroencephalography(EEG) of frontotemporal dementia(FTD), alzheimer`s disease(AD) patients and healthy control(HC) were classified using convolutional neural network(CNN) and evaluated its performancesMethod: EEG dataset containing 88 subjects is downsampled and sliced into 10 seconds. Total dataset was divided into train, test and validation sets for each subject. CNN structure was used to extract temporal and spatial features from EEG and classify patients into 3 groups(FTD/AD/HC). Cross entropy loss was calculated to train the model using adam optimizer. Classification performances were measured to validate model such as accuracy, confusion matrix, AUC(area under curve), F1 Score.Result: After training, validation accuracy was 95.35% and macro‐averaged AUC was 0.9962. F1 score was 0.9469 and the minimum validation accuracy measured among subjects was 57.14% in FTD group. For AD group average accuracy was 92.79%, and for FTD group average accuracy was 97.52%. Average accuracy of healthy control group was 95.26%.Conclusion: We implemented CNN model that can classify EEG of dementia patients with very high performances without expert‐designed feature engineering process. These results imply that deep learning based approach is most promising for EEG‐based patient classification.