Accurate voluntary feed intake (VFI) prediction is critical to the productivity and profitability of ruminant livestock production systems. Simple empirical models have been used to predict VFI for decades, but they are inflexible, restrictive, and poorly accommodate many feeding conditions, such as those of developing countries. We have developed a mechanistic model to predict VFI over a range of forage diets (low- and high-quality grasses and legumes) by wild and domestic ruminants of varying physiological states (growth, lactation, gestation, nonproductive). Based on chemical reactor theory, the model represents the reticulorumen, large intestine, and blood plasma as continuous stirred-tank reactors and the small intestine as a plug flow reactor. Predicted VFI is that which 1) fulfills an empirical relationship between chemostatic and distention feedback observed in the literature, and 2) leads to steady-state conditions. Agreement between observed and actual VFI was great (generally R(2) >0.9, root mean square prediction error <1.4 kg/d, CV <25%). Root mean square prediction error for our model was only 67% that of the Beef NRC (2000) model, the leading empirical prediction system for cattle. Together, these results demonstrate that our model can predict ruminant VFI more broadly and accurately than prior methods and, by consequence, serve as a crucial tool to ruminant livestock production systems.