Mechanism design models typically conclude by characterizing an optimal allocation schedule based on the principal's beliefs regarding agent value functions and the distribution of agent types. This article addresses the question of how a principal can develop these beliefs given a standard cross-sectional data set in which agents' input-output choices are observable, but their underlying heterogeneity is not. I employ the methodology to evaluate strategies for reducing the cost of a voluntary program that reduces cultivation on environmentally-sensitive farmland.