Economists estimating demand and supply systems face the question of using shares or quantities as dependent variables. This article finds that inconsistent estimates are obtained if one makes the wrong choice. A robust structure is presented to let the data choose the preferred form. Empirical applications to U.S. and Texas agriculture using the generalized method of moments suggest that shares and quantities are rejected in favor of a more general functional form. The Texas application provides some evidence for fixed capital but not for labor.