An econometric model of annual per capita U.S. food and nutrition demand is developed. The model is a flexible, full rank two Gorman polar form. It is strictly aggregable across income, demographic variables, and variations in micro preference parameters. Parametric conditions for global quasi-concavity of the (quasi-)utility function are derived. The model is implemented with annual time series data on U.S. per capita food consumption for the sample period 1918-1994. A battery of new test statistics are developed for and applied to the following hypotheses: (1) strict exogeneity of income or total expenditures; (2) global symmetry and negative semidefiniteness of the Slutsky substitution matrix; (3) parameter stability in a multivariate, nonlinear regression model based on within sample residuals; and (4) weak separability of food items from all other goods in the representative consumer's preference function. The empirical results are very encouraging with respect to the strictures of economic theory, heretofore a virtually unheard of outcome. The model is used to analyze the food and nutrient consumption and consumer welfare impacts of the U.S. dairy program.