This paper presents an initial cross-analysis of impact pathways identified in 13 cases studies on agricultural innovation in a developing country context. The aim was to understand better how research actually contributes to impact generation. The results were obtained by applying a common impact assessment approach called ImpresS (Impact of research in the South), which was developed by CIRAD over the past few years. ImpresS includes a set of theory driven, participatory and comprehensive evaluation tools and uses contribution analysis to establish causality links between inputs, outputs, outcomes and impacts. After an overview of the approach used for the cross analysis, we present three case studies, exemplifying different roles researchers play in the generation of impact pathways and especially in that of outputs and outcomes. We then characterize in more generic terms three patterns of impact pathways that are useful for organizing learning. Two key results are outlined as a result of the cross analysis of impact pathways and stakeholders. (1) Research contributes to impact pathways according to three main patterns (2). knowledge across cases can be structured within each of them enabling different kind of learning to inform future research projects.