We explore here our mechanistic understanding of the environmental and physiological processes that determine the oxygen isotope composition of leaf cellulose (δ 18 O cellulose) in a drought-prone, temperate grassland ecosystem. A new allocation-and-growth model was designed and added to an 18 O-enabled soilvegetation-atmosphere transfer model (MuSICA) to predict seasonal (April-October) and multi-annual (2007-2012) variation of δ 18 O cellulose and 18 O-enrichment of leaf cellulose (Δ 18 O cellulose) based on the Barbour-Farquhar model. Modelled δ 18 O cellulose agreed best with observations when integrated over c. 400 growingdegree-days, similar to the average leaf lifespan observed at the site. Over the integration time, air temperature ranged from 7 to 22°C and midday relative humidity from 47 to 73%. Model agreement with observations of δ 18 O cellulose (R 2 = 0.57) and Δ 18 O cellulose (R 2 = 0.74), and their negative relationship with canopy conductance, was improved significantly when both the biochemical 18 O-fractionation between water and substrate for cellulose synthesis (ϵ bio , range 26-30‰) was temperature-sensitive, as previously reported for aquatic plants and heterotrophically grown wheat seedlings, and the proportion of oxygen in cellulose reflecting leaf water 18 O-enrichment (1p ex p x , range 0.23-0.63) was dependent on air relative humidity, as observed in independent controlled experiments with grasses. Understanding physiological information in δ 18 O cellulose requires quantitative knowledge of climatic effects on p ex p x and ϵ bio .