Background: While the importance of conserving ecosystems for sustainable development is widely recognized, it is increasingly evident that despite delivering global benefits, conservation often comes at local cost. Protected areas funded by multilateral lenders have explicit commitments to ensure that those negatively affected are adequately compensated. We make the first comparison of the magnitude and distribution of the local costs of a protected area with the magnitude and distribution of the compensation provided under the World Bank social safeguard policies (Performance Standard 5). Methods: In the Ankeniheny-Zahamena Corridor (a new protected area and REDD+ pilot project in eastern Madagascar), we used choice experiments to estimate local opportunity costs (n = 453) which we annualized using a range of conservative assumptions concerning discount rates. Detailed surveys covering farm inputs and outputs as well as off-farm income (n = 102) allowed us to explore these opportunity costs as a proportion of local incomes. Intensive review of publically available documents provided estimates of the number of households that received safeguard compensation and the amount spent per household. We carried out a contingent valuation exercise with beneficiaries of this compensation two years after the micro-development projects were implemented (n = 62) to estimate their value as perceived by beneficiaries. Results: Conservation restrictions result in very significant costs to forest communities. The median net present value of the opportunity cost across households in all sites was US$2,375. When annualized, these costs represent 27–84% of total annual income for median-income households; significantly higher proportionally for poorer households. Although some households have received compensation, we conservatively estimate that more than 50% of eligible households (3,020 households) have not. Given the magnitude of compensation (based both on amount spent and valuation by recipients two years after the compensation was distributed) relative to costs, we argue that no one was fully compensated. Achieving full compensation will require an order of magnitude more than was spent but we suggest that this should be affordable given the global value of forest conservation. Discussion: By analyzing in unprecedented depth both the local costs of conservation, and the compensation distributed under donor policies, we demonstrate that despite well-intentioned policies, some of the poorest people on the planet are still bearing the cost of forest conservation. Unless significant extra funding is provided by the global beneficiaries of conservation, donors’ social safeguarding requirements will not be met, and forest conservation in developing countries will jeopardize, rather than contribute to, sustainable development goals.