The role of peatlands in the global carbon cycle is confounded by two inconsistencies. First, peatlands have been a large reservoir for carbon sequestered in the past, but may be either net sources or net sinks at present. Second, long–term rates of peat accumulation (and hence carbon sequestration) are surprisingly steady, despite great variability in the short–term rates of peat formation. Here, we present a feedback mechanism that can explain how fine–scale and short–term variability in peat–forming processes is constrained to give steady rates of peat accumulation over longer time–scales. The feedback mechanism depends on a humpbacked relationship between the rate of peat formation and the thickness of the aerobic surface layer (the acrotelm), such that individual microforms (hummocks, lawns, hollows and pools) expand or contract vertically in response to fluctuations in the position of the water table. Hummocks (but not hollows) ‘evolve’ to a steady state where changes in acrotelm thickness compensate for climate–mediated variations in surface wetness. With long–term growth of a topographically confined peat deposit, the steady state gradually shifts to a thicker acrotelm (i.e. taller hummocks) and lower rates of peat formation and carbon sequestration.