The large-scale conversion of N ₂ and H ₂ into NH ₃ (refs. ^1,2) over Fe and Ru catalysts ³ for fertilizer production occurs through the Haber–Bosch process, which has been considered the most important scientific invention of the twentieth century ⁴. The active component of the catalyst enabling the conversion was variously considered to be the oxide ⁵, nitride ², metallic phase or surface nitride ⁶, and the rate-limiting step has been associated with N ₂ dissociation ^7–9, reaction of the adsorbed nitrogen ¹⁰ and also NH ₃ desorption ¹¹. This range of views reflects that the Haber–Bosch process operates at high temperatures and pressures, whereas surface-sensitive techniques that might differentiate between different mechanistic proposals require vacuum conditions. Mechanistic studies have accordingly long been limited to theoretical calculations ¹². Here we use X-ray photoelectron spectroscopy—capable of revealing the chemical state of catalytic surfaces and recently adapted to operando investigations ¹³ of methanol ¹⁴ and Fischer–Tropsch synthesis ¹⁵—to determine the surface composition of Fe and Ru catalysts during NH ₃ production at pressures up to 1 bar and temperatures as high as 723 K. We find that, although flat and stepped Fe surfaces and Ru single-crystal surfaces all remain metallic, the latter are almost adsorbate free, whereas Fe catalysts retain a small amount of adsorbed N and develop at lower temperatures high amine (NH _x) coverages on the stepped surfaces. These observations indicate that the rate-limiting step on Ru is always N ₂ dissociation. On Fe catalysts, by contrast and as predicted by theory ¹⁶, hydrogenation of adsorbed N atoms is less efficient to the extent that the rate-limiting step switches following temperature lowering from N ₂ dissociation to the hydrogenation of surface species.