A fundamental challenge in impact evaluations that rely on a quasi-experimental design is to define a control group that accurately reflects the counterfactual situation. Our aim is to evaluate empirically the performance of a range of approaches that are widely used in economic research. In particular, we compared three different types of matching algorithms (optimal, greedy and nonparametric). These techniques were applied in the evaluation of the impact of the MARENA programme (Manejo de Recursos Naturales en Cuencas Prioritarias), a natural resource management programme implemented in Honduras between 2004 and 2008. The key findings are: (a) optimal matching did not produce better-balanced matches than greedy matching; and (b) programme impact calculated from nonparametric matching regressions, such as kernel or local linear regressions, yielded more consistent outcomes. Our impact results are similar to those previously reported in the literature, and we can conclude that the MARENA programme had a significant, positive impact on beneficiaries.