•An agent based dairy farm model for pasture base system.•Tight interaction between the animal and the grass characteristic.•Take into account the impact of the grass height on the intake of the animals.•Evaluated on a French and Irish farm in term of milk yield and grass utilisation.•Possibility to simulate a wide range of pasture based dairy system. Modelling pasture-based systems is a challenge for modellers worldwide. However, models can play a vital role as grazing management tools and help the decision making process at farm level. The objective of this paper is to describe and evaluate the pasture-based herd dynamic milk (PBHDM) model. The PBHDM model comprises the herd dynamic milk (HDM) model and integrates it with a grazing management and a paddock sub-model. Animal intake at grazing is dependent on the animal characteristics but also on grass availability and quality. It also depends on the interactions between the animal and the grass during the defoliation process. Management of grass on farm can be regulated through different rules during the grazing season including the decision to cut some paddocks in the case of a grass surplus and to allocate supplementation in the case of a grass deficit. The PBHDM was evaluated by comparing model outputs with two grazing systems one in France and one in Ireland. For both farms the grazing season is longer than 7 months. Model outputs that were compared to the actual experimental data included milk production, pre- and post-grazing height and feed supplementation levels. These outputs were all compared on a weekly basis while paddock residence time and total grass harvested as conserved grass silage was evaluated over the grazing season as a whole. The model was capable of reproducing the two grazing systems with acceptable accuracy. It simulated the pre- and post-grazing height with a maximal difference between the actual and the simulated average height through the year of 0.4cm. The model has a tendency to slightly over-estimate the milk production especially in autumn. However in general the model is relatively accurate with a root mean square error less than 20% for the simulated farms.