The principles of mathematical modeling in agricultural sciences are well described by France and Thornley (1984). They categorized models as static or dynamic, empirical or mechanistic, and deterministic or stochastic, although, in practice, they can fall somewhere in the middle of each. In general, our nutrient requirement models are static, empirical, and deterministic; they provide snapshots in time, do not describe the mechanisms underlying responses, and do not consider the inherent variance intrinsic to biological systems. These models are generally easier to derive, and have served the community very well for more than a century. The Molly cow model is dynamic, mechanistic, and deterministic; it predicts responses through time, is based on the underlying driving elements of digestion and metabolism, but does not represent the biological variation underlying predictions. Dynamic models are very useful when one needs to predict changes over time as compared with representing only the new state after the system is given sufficient time to reach steady state. For example, growth and lactation models are typically dynamic, empirical, and deterministic. They capture the effects of slightly greater growth rates on body weight at any point in the growth cycle, or the effect of greater persistency on overall lactational yield. Static nutrient response models only provide the new rate of growth or milk yield after the animal has consumed the diet long enough to reach a new steady state. They cannot predict full lactation yields. Mechanistic models are often used to represent the effects of underlying behavior on higher level performance, e.g., the effects of passage rate on ruminal digestion or the effects of enzymatic activity of a tissue on metabolism. Such representations may provide more precise predictions of higher level performance, although that generally requires that the mechanisms are well defined and provide unbiased estimates. The models are also very useful to assess the relative importance of more basic information. Addition of stochastic elements to mechanistic models can accommodate known variance in the underlying mechanisms and thus provide confidence intervals for predictions.