Carbon (C) stocks in forest soils were evaluated in the first comprehensive survey of Great Britain, the BioSoil soil survey, using a total of 167 plots (72 in England, 26 in Wales and 69 in Scotland). The average C stock down to 80 cm depth for seven main soil types ranged between 108 and 448 t C/ha with maximum values from 511 to 927 t C/ha. Carbon stock varied with soil depth and type, forest type, and stand age. Stocks within the upper mineral soil (0–20 cm) represented between 29 and 69% of the total 0–80 cm C stock, while those in the top 40 cm comprised 59–100% of the total. Carbon stocks decreased in the order deep peats > peaty gleys > groundwater gleys > surface-water gleys > podzols and ironpans > brown earths > rankers and rendzinas. Litter and fermentation horizons on average contributed an additional 7.3 and 8.8 t C/ha, respectively, to the overall soil C stock. Measured soil C stocks (0–80 cm) were upscaled by area of main soil and forest types to provide national estimates. Total forest soil stocks for England, Wales and Scotland were upscaled to 163, 46 and 337 Mt C, respectively, with an additional 17, 4 and 21 Mt C within surface organic layers (litter and fermentation horizons). Carbon stocks were larger under conifers compared with broadleaves. Peaty gleys contributed most to the total C stock in Scotland, while brown earths and podzolic soils made the largest contribution in Wales, and brown earths and surface-water gley soils in England. Estimated total carbon stocks in forest soils in Great Britain, including organic layers, are 589 Mt C in the top 80 cm and 664 Mt C in the top 1 m of soil. The BioSoil soil survey provides the most comprehensive estimate of forest soil C stocks in Great Britain to date and provides a good baseline for assessing future change even though variability in forest soil C stocks is high. However, a relatively small number of additional plots to fill existing gaps in spatial coverage and to increase representation of rendzinas and highly organic soils would significantly reduce the level of uncertainty.