Changes in global soil carbon stocks have considerable potential to influence the course of future climate change. However, a portion of soil organic carbon (SOC) has a very long residence time (> 100 years) and may not contribute significantly to terrestrial greenhouse gas emissions during the next century. The size of this persistent SOC reservoir is presumed to be large. Consequently, it is a key parameter required for the initialization of SOC dynamics in ecosystem and Earth system models, but there is considerable uncertainty in the methods used to quantify it. Thermal analysis methods provide cost-effective information on SOC thermal stability that has been shown to be qualitatively related to SOC biogeochemical stability. The objective of this work was to build the first quantitative thermal analysis based model of the size of the centennially persistent SOC pool. We used a unique set of soil samples from four agronomic experiments in Northwestern Europe with long-term bare fallow and non-bare fallow treatments (e.g. manure amendment, cropland and grassland), as a sample set for which estimating the size of the centennially persistent SOC pool is relatively straightforward. At each experimental site, we estimated the average concentration of centennially persistent SOC and its uncertainty by applying a Bayesian curve fitting method on the observed declining SOC concentration over the duration of the long-term bare fallow treatment. Overall, the estimated concentrations of centennially persistent SOC ranged from 5 to 11 gC.kg−1 soil (lowest and highest boundaries of four 95 % confidence intervals). Then, by dividing site-specific concentrations of persistent SOC by the total SOC concentration of 118 archived soil samples from long-term bare fallow and non-bare fallow treatments, we could estimate the proportion of centennially persistent SOC in the samples and the associated uncertainty. The proportion of centennially persistent SOC ranged from 0.14 (standard deviation of 0.01) to 1 (standard deviation of 0.15). Samples were subjected to thermal analysis by Rock-Eval 6 that generated a series of 30 parameters reflecting their SOC thermal stability and bulk chemistry. The sample set was split into a calibration set (n = 88) and a validation set (n = 30). We trained a non-parametric machine learning algorithm (random forests multivariate regression model) that accurately predicted the size of the centennially persistent SOC pool using Rock-Eval 6 thermal parameters as predictors in the calibration set (pseudo-R² = 0.91, RMSEC = 0.06) and the validation set (R² = 0.91, RMSEP = 0.07). The uncertainty of the predictions obtained using the multivariate regression model was quantified by a Monte Carlo approach that produced conservative 95% prediction intervals across the 30 samples of the validation set. This model based on Rock-Eval 6 thermal analysis can thus be used to predict the proportion of centennially persistent SOC with known uncertainty in new soil samples from similar pedoclimates. Our study strengthens the evidence for a link between the thermal and biogeochemical stability of soil organic matter, and demonstrates that Rock-Eval 6 thermal analysis can be used to quantify the size of the centennially persistent organic carbon pool in temperate soils.