Extensive forest restoration is a key strategy to meet nature-based sustainable development goals and provide multiple social and environmental benefits. Yet achieving forest restoration at scale requires cost-effective methods. Tree planting in degraded landscapes is a popular but costly forest restoration method, which often results in less biodiverse forests when compared to natural regeneration techniques under similar conditions. Here, we assess the current spatial distribution of pantropical natural forest (from 2000-2016) and use this information to present the first model of the potential for natural regeneration across tropical forested countries and biomes at 30-meter spatial resolution. We estimate that 215 million hectares - an area greater than the entire country of Mexico - have potential for natural forest regeneration, representing an above-ground carbon sequestration potential of 23.4 Gt CO2 (range 21.1-25.7 Gt) over 30 years. Five countries (Brazil, Indonesia, China, Mexico, and Colombia) account for 52% of this estimated potential, showcasing the need for targeting restoration initiatives that leverage natural regeneration potential. Our results facilitate broader equitable decision-making processes that capitalise on the widespread opportunity for natural regeneration to help achieve national and global environmental agendas. File descriptions: pnv_pct_30m - files named with this prefix are the 30m continuous probability predictions of the potential for natural regeneration, stored as integers representing percentages to minimise file sizes tile_index_number_map.jpg – corresponding locations of the ‘pnv_pct_30m’ tiles prop pnv v1 1km.tif - for display purposes only (not used in the Williams et al. analysis) the approximately 1km resolution overview raster representing the proportion of the area of each pixel that has the potential for natural regeneration pnv_bin_30m.zip - this .zip folder contains the 30m resolution binary predictions (where a value of >0.5 is allocated a value of 1, acknowledging users may prefer to use a different threshold for their context) of areas suitable for natural regeneration The 30m resolution datasets have been tiled into 10 degree latitude/longitude tiles