Background: The present analysis revisit the impact of extremely low-frequency magnetic fields (ELF MF) on melatonin (MLT) levels in human and rat subjects using both a parametric and non-parametric approach. Method: In this analysis, we use 62 studies from review articles. The parametric approach consists in a Bayesian Logistic Regression (LR) analysis and the non-parametric approach consists of a Support Vector analysis which are designed to be robust against spurious/false results. Results: Both approach reveal a unique well ordered pattern, and show that humans and rat studies are consistent with each other once the MF strength is restricted to cover the same range (with B 50µT). In addition, the data reveal that chronic exposure (longer than ∼ 22 days) to ELF MF appears to decrease MLT levels only when the MF strength is below a threshold of 30µT (log B thr /µT= 1.4 +0.7 −0.5), i.e. when the man-made ELF MF intensity is below that of the static geomagnetic field. Conclusions: Studies reporting an association between ELF MF and changes to MLT levels and some the opposite (no association with ELF MF) can be reconciled under a single framework.