Assessing the performance of efforts to reduce emissions from deforestation and forest degradation (REDD+) requires data on forest cover change. Innovations in remote sensing and forest monitoring provide ever-increasing levels of coverage, spatial and temporal detail, and accuracy. More global products and advanced open-source algorithms are becoming available. Still, these datasets and tools are not always consistent or complementary, and their suitability for local REDD+ performance assessments remains unclear. These assessments should, ideally, be free of any confounding factors, but performance estimates are affected by data uncertainties in unknown ways. Here, we analyse (1) differences in accuracy between datasets of forest cover change; (2) if and how combinations of datasets can increase accuracy; and we demonstrate (3) the effect of (not) doing accuracy assessments for REDD+ performance measurements. Our study covers five local REDD+ initiatives in four countries across the tropics. We compared accuracies of a readily available global forest cover change dataset and a locally modifiable open-source break detection algorithm. We applied human interpretation validation tools using Landsat Time Series data and high-resolution optical imagery. Next, we assessed whether and how combining different datasets can increase accuracies using several combination strategies. Finally, we demonstrated the consequences of using the input datasets for REDD+ performance assessments with and without considering their accuracies and uncertainties. Estimating the amount of deforestation using validation samples could substantially reduce uncertainty in REDD+ performance assessments. We found that the accuracies of the various data sources differ at site level, although on average neither one of the input products consistently excelled in accuracy. Using a combination of both products as stratification for area estimation and validated with a sample of high-resolution data seems promising. In these combined products, the expected trade-offs in accuracies across change classes (before, after, no change) and across accuracy types (user’s and producer’s accuracy) were negligible, so their use is advantageous over single-source datasets. More locally calibrated wall-to-wall products should be developed to make them more useful and applicable for REDD+ purposes. The direction and degree of REDD+ performance remained statistically uncertain, as CIs were overlapping in most cases for the deforestation estimates before and after the start of the REDD+ interventions. Given these uncertainties and inaccuracies and to increase the credibility of REDD+ it is advised to (1) be conservative in REDD+ accounting, and (2) not to rely on results from single currently available global data sources or tools without sample-based validation if results-based payments are intended to be made on this basis.