Mammalian body, blood and hard tissue oxygen isotope compositions (δ18O values) reflect environmental water and food sources, climate, and physiological processes. For this reason, fossil and archaeological hard tissues, which originally formed in equilibrium with body chemistry, are a valuable record of past climate, landscape paleoecology, and animal physiology and behavior. However, the environmental and physiological determinants of blood oxygen isotope composition have not been determined experimentally from large herbivores. This class of fauna is abundant in Cenozoic terrestrial fossil assemblages, and the isotopic composition of large herbivore teeth has been central to a number of climate and ecological reconstructions. Furthermore, existing models predict blood water, or nearly equivalently body water, δ18O values based on environmental water sources. These have been evaluated on gross timescales, but have not been employed to track seasonal variation. Here we report how water, food, and physiology determine blood water δ18O values in experimental sheep (Ovis aries) subjected to controlled water switches. We find that blood water δ18O values rapidly reach steady state with environmental drinking water and reflect transient events including weaning, seasons, and snowstorms. Behavioral and physiological variation within a single genetically homogenous population of herbivores results in significant inter-animal variation in blood water δ18O values at single collection times (1 s.d. = 0.1–1.4 ‰, range = 3.5 ‰) and reveals a range of water flux rates (t1/2 = 2.2–2.9 days) within the population. We find that extant models can predict average observed sheep blood δ18O values with striking fidelity, but predict a pattern of seasonal variation exactly opposite of that observed in our population for which water input variation was controlled and the effect of physiology was more directly observed. We introduce to these models an evaporative loss term that is a function of environmental temperatures. The inclusion of this function produces model predictions that mimic the observed seasonal fluctuations and match observations to within 1.0 ‰. These results increase the applicability of available physiological models for paleoseasonality reconstructions from stable isotope measurements in fossil or archaeological enamel, the composition of which is determined in equilibrium with blood values. However, significant blood δ18O variation in this experimentally controlled population should promote caution when interpreting isotopic variation in the archaeological and paleontological record.