A method is proposed for testing the robustness of the assessed targeting performance of an anti-poverty program to the fact that program administrators have a broader concept of "poverty" than the economist/evaluator. An application is given to China's main urban anti-poverty program.