The visual evoked potential as an electrophysiological signal is mainly used in the neurophysiological exploration of the optic nerves. Traditionally, medical doctors base their diagnosis of specific pathologies related to the time delay of the nerve flow on the time scale. In this context, the VEP latency P100 that reflects a temporal notion is considered the main characteristic on which human interpretation is based. However, its value is influenced by different factors and remains a limited method. This insufficiency triggers our interest instead in deep learning architectures, taking into consideration and adapting to the specificity of each particularity related to the laboratory of the neurophysiological exploration unit in the hospital. The comparison between the results obtained from Matlab by the application of the CNN as well as the RNN, based on the evaluation parameters calculated after k-fold cross-validation, confirms that the CNN-1D architecture can be considered powerful in terms of reliability of classification between signals that are related to pathological subjects and normal ones, which privileges the use of this architecture compared with recurrent neural networks that are less reliable and require more time for execution, subsequently the use of the CNN will allow us to avoid even the extraction of attributes for the discrimination between the two classes object of classification, with the possibility to progressively improve the performance of the solution over time based on the new signals acquired in the VEP analysis laboratory.