In this manuscript, we analyze different topologies of Convolutional Neural Networks (CNN) for classifying the P300 wave from an EEG signal. Also, we propose a selection criteria in order to improve the classification accuracy. In this study, the brain signals of healthy and disabled subjects were analyzed and four architectures were tested with different numbers of filters with the same dimensions. The results of the current work indicate that the best bitrate in disabled and healthy subjects was 14.14 and 25.44 bits per minute, respectively. Using target by block evaluation, the classification accuracy of 100% was obtained in healthy and disabled subjects. This approach is compared to various machine learning algorithms so that our results outperformed others works.