Nitrogenase is the only enzyme that can cleave the strong triple bond in N 2 , making nitrogen available for biological lifeforms. The active site is a MoFe 7 S 9 C cluster (the FeMo cluster) that binds eight electrons and protons during one catalytic cycle, giving rise to eight intermediate states E 0 -E 7 . It is experimentally known that N 2 binds to the E 4 state and that H 2 is a compulsory byproduct of the reaction. However, formation of H 2 is also an unproductive side reaction that should be avoided, especially in the early steps of the reaction mechanism (E 2 and E 3 ). Here, we study the formation of H 2 for various structural interpretations of the E 2 -E 4 states using combined quantum mechanical and molecular mechanical (QM/MM) calculations and four different density-functional theory methods. We find large differences in the predictions of the different methods. B3LYP strongly favours protonation of the central carbide ion and H 2 cannot form from such structures. On the other hand, with TPSS, r 2 SCAN and TPSSh, H 2 formation is strongly exothermic for all structures and E n and therefore need strict kinetic control to be avoided. For the E 2 state, the kinetic barriers for the low-energy structures are high enough to avoid H 2 formation. However, for both the E 3 and E 4 states, all three methods predict that the best structure has two hydride ions bridging the same pair of Fe ions (Fe2 and Fe6) and these two ions can combine to form H 2 with an activation barrier of only 29-57 kJ mol -1 , corresponding to rates of 7 × 10 2 to 5 × 10 7 s -1 , i.e. much faster than the turnover rate of the enzyme (1-5 s -1 ). We have also studied H-atom movements within the FeMo cluster, showing that the various protonation states can quite freely be interconverted (activation barriers of 12-69 kJ mol -1 ).