To enhance understanding of environmental science–policy interactions, this study analyses how environmental knowledge is produced, circulated, and applied in the Reduced Emissions from Deforestation and Degradation (REDD +) programme in Zambia. Data are drawn from interviews with key actors in the REDD + process and an extensive critical review of policy documents and deforestation estimates. We find that research over the past 50 years has not resulted in accurate estimates of forest cover and deforestation rates, nor have major deforestation drivers been convincingly documented. Estimates are difficult to compare due to inconsistent use of key terms, methodological pluralism and differences in social framing. We argue that an epistemic community is able to influence production, circulation, and application of deforestation related knowledge. Furthermore, in a situation of weak and contradictory empirical evidence, this community is arguably able to sustain a deforestation discourse centred on high forest loss and neo-Malthusian causal explanations. This is done through mechanisms making it difficult to separate facts from politics, e.g. by black boxing the origin and units of measure of deforestation estimates. We argue that this makes it more difficult to realise positive outcomes through REDD + implementation.