Projects for reducing emissions from deforestation and forest degradation (REDD+) have been initiated in developing countries, featuring partnerships with multiple actors under the global forest and climate change regime. Even though partnerships between stakeholders are crucial for ensuring successful project deliveries, there is a lack of knowledge about sectoral partnerships within and between stakeholders in REDD+ projects. This study aims to measure the structures and patterns of REDD+ project partnerships using an original, multi-stage social network theory approach with global- and regional-level centralization analyses using three major regions (Asia, Africa and South America), and configurations using exponential random graph modeling (ERGM). Using data on 480 REDD+ projects implemented in 57 countries, results show concentrated polycentric networks across several dominant actors, including USA-, Brazil- and China-based organizations. Statistical network modeling indicates that, overall, partnerships are less likely to be created between different organization categories (across-type bridging), but tend more towards cooperation with the same types (within-type bridging). Research institutes, however, produce distinctly different patterns, forming across-type partnerships with highly technical capacities. Comparisons of stakeholders at different stages of the REDD+ mechanism help in understanding the complete picture of REDD+ architecture. This study contributes by offering insights for designing future partnerships within REDD+ projects and suggests ways to improve multi-level collaboration and cooperation.