EEG-based biometric identification has been investigated for some time. Recently, a challenge on EEG-person-identification at ICASSP2025 yielded remarkable results, with 8 teams reaching 100% accuracy. However, it remains unclear whether the models truly learn biometric features or rely on system-induced biases, such as electrode layout. We investigate the impact of electrode arrangement, brain areas, and frequency bands on EEG-based person identification. Our findings show that deep learning models often hinge on electrode placement, yet training on varied electrode subsets preserves classification performance, indicating extraction of genuine biometric cues. Moreover, certain brain regions and frequency bands exhibit stronger discriminative characteristics. These results underscore the combined influence of spatial distribution and subject-specific EEG features, emphasizing the need to consider spatial and spectral factors for robust EEG-based biometrics and neuroscience research.