Growing awareness of the multiple environmental impacts of livestock production has created a need to extend the definition of efficiency to a multidimensional eco-efficiency concept. Our paper proposes an original methodology, named “simulation based DEA” that combines an agronomic whole-farm model and an efficient frontier method for in-depth exploration of the determinants of eco-efficiency. GAMEDE, the whole-farm model we use, is randomly parameterized for key management practices and structural parameters to generate a large dataset of simulated dairy systems. The upper and lower bounds set for the stochastic choice of parameter values are a key point in the methodology and rely on expert knowledge derived from participatory modelling. For each simulation, numerous indicators describing the functioning and the performance of the production system are calculated and a set of inputs, outputs, and undesirable outputs, are then used to define the production technology in an efficiency frontier analysis. Data envelopment analysis, the efficiency frontier method applied, provides a multidimensional eco-efficiency score representing the increase in outputs that is possible with no increase in inputs or undesirable outputs. The eco-efficiency score can be linked to all the indicators of the production system calculated by the whole-farm model, after which it becomes possible to explore the managerial, structural, economic, agronomic, zootechnical, and environmental factors that explain different levels of eco-efficiency. In our case study, dairy farming in Reunion Island, livestock production is strongly constrained by land scarcity. Consequently the most eco-efficient farms appear to be intensive systems with high forage productivity to ensure feed self-sufficiency. While most studies on efficiency are restricted to a narrow dataset, the proposed methodology is innovative in that it makes it possible to cover a wide range of possible livestock farming systems in a giv