Interest in the economics of plant breeding first emerged during the late 1960s and early 1970s following the well-known green revolutions in wheat and rice. Since that time, few branches of agricultural research have been subjected to as much scrutiny as plant breeding. Impacts assessment studies consistently conclude that the economic benefits generated by successful plant breeding programs are large, positive, and widely distributed. Case studies repeatedly find that investment in crop genetic improvement generates attractive rates of return compared to alternative investment opportunities. Similarly, case studies consistently show that the welfare benefits resulting from the adoption of modern varieties (MVs) are broadly shared by producers and consumers in both favored and marginal environments. Swayed by the large body of empirical evidence that supports these findings, governments, lending agencies, philanthropic organizations, and private corporations have invested millions in plant breeding research. But how reliable are the results of studies that estimate the benefits of plant breeding research? Are the methods used to conduct such studies theoretically sound? And are the data sufficiently complete and accurate? This paper reviews methods used to estimate the benefits of plant breeding research and discusses methodological issues and practical challenges that often receive inadequate attention in applied impacts work. Our goal is not to question the validity of the broad conceptual frameworks used to estimate the benefits of plant breeding research (e.g., economic surplus approach, production function approach). Nor is it our intention to examine theoretical issues that complicate research evaluation in general. Rather, the objectives of the paper are to examine problems that may arise during empirical evaluation of plant breeding research, to discuss what can happen if those problems are ignored or overlooked, and to propose workable solutions. The problems that affect the empirical evaluation of plant breeding research can be grouped into three general categories: measuring the adoption of MVs; estimating the benefits associated with adoption of MVs; and assigning credit for plant breeding research. In principle, estimating the area planted to MVs should be relatively easy. In practice, it is often very difficult. Important issues in estimating the benefits associated with MV adoption include the relationship between experimental and farm level yields; the relative effects of improved germplasm and improved crop management practices; maintenance research; non-yield benefits; and constructing counterfactual scenarios. The existence of spillovers in plant breeding research increases the overall benefits generated by the global plant breeding system, but it also complicates the task of assigning credit among individual breeding programs. CIMMYT researchers, working in collaboration with colleagues from national agricultural research organizations, have conducted a series of studies designed to document and quantify the impacts of international maize and wheat breeding research. Drawing on lessons learned from the CIMMYT studies, each of these three sets of problems that can affect the empirical evaluation of plant breeding research are discussed in detail, and practical guidelines are presented to help those interested in conducting applied impacts studies avoid common pitfalls that if ignored may lead to incorrect empirical results.