The grazing process determines not only the nutrient intake of ruminants at pasture but also the intensity of their impact on vegetation. Grazing dynamics are the result of complex interactions between animal and sward characteristics. While sward is being depleted, a decline in intake rate is partly offset by longer grazing time. This behaviour may be controlled by nutritional feedback from digestion and nutrient absorption as the quality of the ingested herbage decreases. To simulate the dynamics of feeding behaviour and intake during sward exploitation, we developed a mechanistic model of intake rate that combines sward architecture and foraging decisions, and we linked this model with another focusing on control of intake. The sward was divided into horizons characterised by bulk density and nutritive value (NDF content and digestibility), enabling prediction of bite mass and potential intake rate for each grazed horizon. Animal decisions were simulated at two levels: (i) animal activity (eating, ruminating or resting) was self-regulated every minute by comparing a motivation-to-eat function with a satiation function based on a digestion and metabolic sub-model; (ii) while eating, the horizon to be grazed was decided through a choice function taking into account the relative availabilities and potential intake rates of the two upper horizons. The model simulates the animal-sward interactions from elementary parameters (bite mass, intake rate, etc.) to integrated outputs (sward height, daily intake, etc.). Hence, the interplay between characteristics of the vegetation and the internal state of the animal is dynamic taken from the level of a few bites to several successive days. Satisfactory validations were obtained on experimental data sets obtained in both rotational grazing with dry ewes and continuous grazing with lactating ewes. The sensitivity analysis highlights the balance between factors that control bite mass and intake rate and factors that control grazing time. Combining both series of factors in the same model represents significant progress in predicting intake at grazing in a mechanistic way. The model makes it possible to explore the balance between intake regulation by nutritional variables (animal needs and sward quality) and by the availability and structure of the sward. Finally, it is a promising tool to explore the sensitivity of the grazing process to characteristics of both sward (height, bulk density, nutritive value) and animals (weight, nutritional requirements, behavioural traits), and to management practices (stocking rate, rotational versus continuous grazing).