Inaccurate estimate of the largest terrestrial carbon pool, soil organic carbon (SOC) stock, is the major source of uncertainty in simulating feedback of climate warming on ecosystematmosphere carbon exchange by process based ecosystem and soil carbon models. Although the models need to simplify complex environmental processes of soil carbon sequestration, in a large mosaic of environments a missing key driver could lead into a modelling bias in predictions of SOC stock change. We aimed to evaluate SOC stock estimates of process based models (Yasso07, Q, and CENTURY) against the Swedish forest soil inventory data (3230 samples) organized by recursive partitioning method into distinct soil groups with underlying SOC stock development linked to physicochemical conditions. The Yasso07 and Q models that used only climate and litterfall input data and ignored soil properties generally agreed with two third of measurements. However, in fertile sites with high nitrogen deposition, high cation exchange capacity, or moderately increased soil water content, Yasso07 and Q underestimated SOC stocks. Accounting for soil texture (clay, silt, and sand content) and structure (bulk density) in CENTURY model showed no improvement on carbon stock estimates, as CENTURY deviated in similar manner. Our analysis suggested that the soils with poorly predicted SOC stocks, as characterized by the high nutrient status and well sorted parent material, indeed have had other predominat drivers of SOC stabilization lacking in the models presumably the mycorrhizal organic uptake and organo-mineral stabilization processes. Our results imply that the role of soil nutrient status as regulator of organic matter mineralization has to be re-evaluated, since correct steady state SOC stocks are decisive for predicting future SOC change.