We integrated two methods, experimental heating and observations across natural climate gradients, to elucidate both short- and long-term climatic controls on ecosystem carbon storage and to investigate carbon-cycle feedbacks to climate in montane meadows. A 10-year heating experiment warmed and dried heated plot soils and substantially decreased (by ~200 ¿ 150 g C m-2) the amount of carbon stored in soil organic matter, a positive feedback to warming. In situ CO2 flux measurements, laboratory soil incubations, and a heating-induced shift in vegetation community composition from high- to low-productivity species indicate that a decline in community productivity and resultant decrease in soil inputs from plant litter caused most of the soil carbon decrease. An alternative widely hypothesized mechanism for soil carbon decrease under warming is stimulation of soil respiration, but we observed no increase in seasonally integrated soil respiration in our experiment (soil drying inhibited microbial decomposition even as soil warming stimulated it). To extend our analysis from the short-term transient response represented by the heating experiment to the presumed long-term approximate steady state represented by natural climate gradients, we tested a hypothesized relation between vegetation community composition (which controls both litter input rate and average litter quality) and soil carbon along the climate gradient. The gradient analysis implies that the experimentally induced decline in soil carbon is transient and will eventually reverse as lower quality litter inputs from the increasingly dominant low-productivity species reduce soil respiration losses. This work shows that ecological processes can control both short- and long-term responses to climate change, confirming some model-based predictions about the importance of vegetation shifts, but challenging the widely held hypothesis that the effect of temperature change on respiration will dominate soil carbon changes.