Benefit-cost analysis consists of establishing ratios of benefits to costs for a set of project variants. The decision rule is to select that project variant where the ratio is a maximum. This paper argues that specification and estimation errors can contribute to findings for large-scale systems of benefit-cost ratios approximating zero. The feasibility of descriptive, structural, and experimental approaches to benefit-cost analysis is discussed. A benefit-cost model is presented for the small-scale educational project. This model is generalized to a large-scale system, and it is demonstrated that scaleup factors such as communication and control in the organization, often overlooked, provide specification errors that contribute to the finding of negligible benefit-cost ratios. (Author)