Agricultural development is fundamentally a social process in which people construct solutions to their problems, often by modifying both new technologies and their own production systems to take advantage of new opportunities offered by the technologies. Hence, agricultural change is an immensely complex process, with a high degree of non-linearity. However, current ‘best practice’ economic evaluation methods commonly used in the CGIAR system ignore complexity. In this paper we develop a two-stage monitoring, evaluation and impact assessment approach called impact pathway evaluation. This approach is based on program-theory evaluation from the field of evaluation, and the experience of the German development organization GTZ (Deutsche Gesellschaft für Technische Zusammenarbeit GmbH). In the first stage of this approach, a research project develops an impact pathway for itself, which is an explicit theory or model of how the project sees itself achieving impact. The project then uses the impact pathway to guide project management in complex environments. The impact pathway may evolve, based on learning over time. The second stage is an ex post impact assessment sometime after the project has finished, in which the project's wider benefits are independently assessed. The evaluator seeks to establish plausible links between the project outputs and developmental changes, such as poverty alleviation. We illustrate the usefulness of impact pathway evaluation through examples from Nigeria and Indonesia.