In this study, we explore the feasibility of using a convolutional neural network (CNN) to decode human electroencephalographic (EEG) response for user authentication. In particular, we exploit the low-frequency components of the steady-state visual-evoked potentials (SSVEP) that contain consistent individualized patterns as the biometric. We evaluate the discriminating capabilities across different parameter configurations to optimize the CNN model. We also investigate how the length of EEG data impact the authentication performance. Our proposed framework achieved ~97% accuracy of crossday user authentication across 8 subjects, shedding light on a practical EEG-based biometric powered by the CNN-based brain decoding.