AbstractThe targeting efficiency and the coverage of social programs for the poor are typically analyzed by partitioning the total population in four mutually exclusive groups: the poor who benefit from a program or policy, the poor who do not benefit, the non‐poor who benefit, and the non‐poor who do not benefit. While useful, this partition into crisp sets may not capture the difficulty of identifying the poor. This paper presents a method that consists of using a membership function to identify to what extent households can be considered as poor or non‐poor. The method builds on fuzzy sets theory whereby the definition of the boundaries of a set, say the poor or the non‐poor, is fuzzy. We characterize the properties that membership functions should have, and we test for the robustness of targeting performance comparisons to the choice of the membership function.