The federal government has estimated the prevalence of household “food insecurity” and “food insecurity with hunger” since 1995. Early observers believed that the new measure could be used to assess and improve the Food Stamp Program (FSP). Ten years of research have tempered the initial optimism. The prevalence of food insecurity with hunger (12.3% of all low-income households in 2004) is much higher among food stamp participant households (18.6% in 2004) than among low-income nonparticipant households (10.1% in 2004), due to strong self-selection effects. Households facing greater hardship are more likely to join the program. This article reviews 6 types of nonexperimental research designs that have been used to address the selfselection problem. The results have been inconclusive and the authors have warned against drawing causal inferences from their research. Ethical random-assignment research designs may be required to satisfy the intense policy interest in measuring the antihunger impact of the FSP. The most promising ethical research designs would test the effects of offering eligibility to households that are currently ineligible or offering increased benefits to households that are currently eligible for small benefit amounts.