Abstract There is growing tendency to elicit subject-dependent features from physiological data like electroencephalogram (EEG) signals, which is the most relative one to the subject’s brain, though the behavior of this signal does not obey any grammatical pattern. In this regard, several attempts have been made toward eliciting subject-dependent features from EEGs to promisingly verify/identify subjects. In this paper, we have comprehensively assessed state-of-the-art EEG features to empirically show their pros and cons. Herein the resting-EEG and ERP-EEG of 20 healthy subjects are utilized and the conventional features are extracted from them. These features are then fed to three types of classifiers. Among the deployed features, spectral coherence and correlation features provide the best verification and identification results in terms of classification accuracy and equal error rate. Empirical results demonstrate that the ERP-EEG features are more discriminative than the resting-EEG features because they reveal the response of a specific circuit of the neural brain system.