This paper examines the cost structure of emissions abatement through different types of Clean Development Mechanism (CDM) projects. Alternative models for abatement costs are specified and estimated using CDM project-specific data. Empirical results indicate that there exist economies of scale in emission abatement through the CDM projects, and that the marginal cost of abatement significantly varies across different types of projects. The distribution of various CDM project types corresponds to the relative attractiveness of the types, in terms of the structure of the estimated marginal cost function. Thus, empirical results suggest that the CDM market operates efficiently and sends the right signals to the investors, which further explains the shying away from costly carbon sequestration projects funded by many international development agencies, such as the World Bank. Contrary to the hypothesis that that the marginal costs of abatement through CDM decrease over time due to experience or learning by doing, empirical results show non-decreasing marginal cost of abatement over time. This finding suggests that there may be other incentives to invest in certain types of CDM projects in specific locations, thus implying location-specificity of various investment opportunities. While non-decreasing marginal cost of abatement over time implies a tougher prospect for CDM in future commitment periods, the current growth pattern of the CDM suggests that this flexibility provision of the Kyoto Protocol is still highly attractive for the host and investor countries.