Considerable efforts have been made in gathering scientific data and developing feeding systems for ruminant animals in the past 50 years. Future endeavours should target the assessment, interpretation and integration of the accumulated knowledge to develop nutrition models in a holistic and pragmatic manner. We highlight some of the areas that need improvement. A fixed metabolisable-to-digestible energy ratio is an oversimplification and does not represent the diversity of existing feedstock, but, at the same time, we must ensure the internal consistency and dependency of the energy system in models. For grazing animals, although data exist to compute energy expenditure associated with walking in different terrains, nutrition models must incorporate the main factors that initiate and control grazing. New equations have been developed to predict microbial crude protein (MCP) production, but efforts must be made to account for the diversity of the rumen microbiome. There is large and unexplained variation in the efficiency of MCP synthesis (9.81-16.3 g MCP/100 g of fermentable organic matter). Given the uncertainties in the determination of MCP, current estimates of metabolisable protein required for maintenance are biased. The use of empirical equations to predict MCP, which, in turn, is used to estimate metabolisable protein intake, is risky because it establishes a dependency between these estimates and creates a specificity that is not appropriate for mechanistic systems. Despite the existence of data and knowledge about the partitioning of retained energy into fat and protein, the prediction of retained protein remains unsatisfactory, and is even less accurate when reported data on the efficiency of use of amino acids are employed in the predictive equations. The integrative approach to develop empirical mechanistic nutrition models has introduced interconnected submodels, which can destabilise the predictability of the model if changed independently.