Globally, soil organic matter (SOM) contains more than three times as much carbon as either the atmosphere or terrestrial vegetation. Yet it remains largely unknown why some SOM persists for millennia whereas other SOM decomposes readily-and this limits our ability to predict how soils will respond to climate change. Recent analytical and experimental advances have demonstrated that molecular structure alone does not control SOM stability: in fact, environmental and biological controls predominate. Here we propose ways to include this understanding in a new generation of experiments and soil carbon models, thereby improving predictions of the SOM response to global warming.Understanding soil biogeochemistry is essential to the stewardship of ecosystem services provided by soils, such as soil fertility (for food, fibre and fuel production), water quality, resistance to erosion and climate mitigation through reduced feedbacks to climate change. Soils store at least three times as much carbon (in SOM) as is found in either the atmosphere or in living plants 1 . This major pool of organic carbon is sensitive to changes in climate or local environment, but how and on what timescale will it respond to such changes? The feedbacks between soil organic carbon and climate are not fully understood, so we are not fully able to answer these questions 2-7 , but we can explore them using numerical models of soil-organic-carbon cycling. We can not only simulate feedbacks between climate change and ecosystems, but also evaluate management options and analyse carbon sequestration and biofuel strategies. These models, however, rest on some assumptions that have been challenged and even disproved by recent research arising from new isotopic, spectroscopic and molecularmarker techniques and long-term field experiments.Here we describe how recent evidence has led to a framework for understanding SOM cycling, and we highlight new approaches that could lead us to a new generation of soil carbon models, which could better reflect observations and inform predictions and policies.