Rigorous impact evaluations of agricultural research and development programs involve estimating the causal effects of interventions. Here we examine the scope and challenges of using an instrumental variable (IV) approach in this connection. To this end, we systematically review 31 impact evaluations conducted in the maize production systems in developing countries. Most of the reviewed studies derive their IVs from cross-sectional survey data. There are some common approaches across the studies to construct IVs, but most violate the underlying assumptions. The proliferation of ‘bad instruments’ questions the validity of impact assessment itself. We acknowledge that finding valid IVs for cross-sectional studies is one of the most daunting tasks in applied economics. Several approaches to find better IVs for impact assessment in agriculture are discussed.