We simulated soil carbon stock dynamics of an Austrian coniferous forest stand with five soil-only models (Q, ROMUL, RothC, SoilCO2/RothC and Yasso07) and three plant-soil models (CENTURY, CoupModel and Forest-DNDC) for an 18-year period and the decomposition of a litter pulse over a 100-year period. The objectives of the study were to assess the consistency in soil carbon estimates applying a multi-model comparison and to present and discuss the sources of uncertainties that create the differences in model results. Additionally, we discuss the applicability of different modelling approaches from the view point of large-scale carbon assessments.Our simulation results showed a wide range in soil carbon stocks and stock change estimates reflecting substantial uncertainties in model estimates. The measured stock change estimate decreased much more than the model predictions. Model results varied not only due to the model structure and applied parameters, but also due to different input information and assumptions applied during the modelling processes. Initialization procedures applied with the models induced large differences among the modelled soil carbon stocks and stock change estimates. Decomposition estimates of the litter pulse driven by model structures and parameters also varied considerably.Our results support the use of relatively simple soil-only models with low data requirements in inventory type of large-scale carbon assessments. It is important that the modelling processes within the national inventories are transparently reported and special emphasis is put on how the models are used, which assumptions are applied and what is the quality of data used both as input and to calibrate the models.