The change in soil carbon (C) stock over a 19–31-year period (mean 25 years) has been measured at 179 sites on a 20-km grid across Scotland. Sampling was by horizon from a profile pit. Although soil bulk density determinations were absent at the first sampling time, we used bulk density values from the second sampling time calibrated against NIR spectra to predict the missing values. There was no detectable change in overall total soil C stock (mean ± standard error, to a depth of 100 cm), which was 266 ± 15 and 270 ± 15 t C ha−1 for the first and second sampling times, respectively, or generally in C stock within specific vegetation or soil types. The exception was for soils under woodland, excluding those on deep peat, which exhibited a significant (P = 0.05) gain of 1.0 t C ha−1 year−1. Soils under woodland (mainly coniferous plantation) also showed a significant (P = 0.04) increase in C content (g kg−1), a significant decrease in bulk density (P = 0.006) and an increase in the thickness of the Litter-Fermentation-Humus (LFH) layer (P = 0.06). Recalculating the C stock to a depth of 15 cm showed a significant increase in overall C stock (when deep peat sites were excluded) as well as specifically in moorland and woodland soils, suggesting that had we sampled only to 15 cm, we would have reached a different conclusion. Both improved grassland soils and those initially under arable cultivation showed a significant decrease in C content. However, the mean thickness of Ap horizons increased from 29 to 32 cm, with a concomitant decrease in C content and a slight increase in bulk density; this we ascribe to deeper ploughing between the sample periods. In the context of possible soil C losses, we can be 95% confident that the mean loss does not exceed 0.2% year−1 and 99% confident that it does not exceed 0.4% year−1.