Most evaluations of the impact of cost sharing policies designed to reduce non-point source pollution on water quality fail to incorporate behavioral responses, which are important because of the voluntary nature of these programs. In this article, a two-stage simultaneous equation approach is applied to data from a farmer survey to correct for voluntary self-selection into cost sharing programs, and account for substitution effects among conservation practices. The estimates obtained from the econometric model are linked with the Chesapeake Bay Program watershed model to estimate the change in abatement levels and marginal abatement costs for nitrogen, phosphorus and sediment after considering non-additional adoption due to nonrandom enrollment, as well as potential indirect effects on other conservation practices. We find that policy scenarios which do not account for non-additional adoption significantly overestimate the abatement achieved by environmental incentive payments. Accounting for nonrandom enrollment increases the average marginal cost of abatement by between 37 and 85 percent across the state of Maryland. However, estimated indirect effects of cost sharing suggest the presence of “crowding in” of other practices, leading to greater abatement and lower costs, particularly for phosphorus and sediment.