The language of economics is the language of models. Understanding the structure of this language offers many benefits. Unfortunately, the structure is ubiquitous in implementation but absent in documentation. This paper documents the structure of models in the context of the theory reduction and testing process. The structure is used to explain why there are several legitimate ways to deal with nonspherical errors in econometric models and why the recent work on stochastic preferences and technologies is a progressive step forward for the discipline. A production modeling exercise is presented to help illuminate the concepts.