Abstract. Predictive, regional use of soil organic matter (SOM) models requires evaluation of the performance of models with datasets from long-term experiments relevant to the scenarios of interest to the regional scale study, and relevant to the climate of the study region. Datasets from six long-term experiments were used to evaluate the performance of RothC and CENTURY, two of the most widely used and tested SOM models. Three types of model run were completed for each site: (1) CENTURY model alone; (2) RothC model run to fit measured SOC values, by iteratively adjusting C inputs to soil; and (3) RothC model run using C inputs derived from CENTURY runs. In general, the performance of both models was good across all datasets. The runs using RothC (iteratively changing C inputs to fit measured SOC values) tended to have the best fit to model data, since this method involved direct fitting to observed data. Carbon inputs estimated by RothC were, in general, lower than those estimated by CENTURY, since SOC in CENTURY tends to turn over faster than SOC in RothC. The runs using RothC with CENTURY C inputs tended to have the poorest fit of all, since CENTURY predicted greater C inputs than were required by RothC to maintain the same SOC content. A plausible model fit to measured SOC data may be obtained with widely differing C input values, due to differences in predicted decomposition rates between models. It remains unclear which, if either, modelling approach most closely represents reality since both C inputs to soil and decomposition rates for bulk SOM are difficult to determine experimentally. Further progress in SOM modelling can only be the result of research leading to better process understanding, both of net C inputs to soil and of SOM decomposition rates. The use of default methods for estimating initial SOC pools in RothC and CENTURY may not always be appropriate and may require adjustment for specific sites. The simulations presented here also suggest details of SOC dynamics not shown by available measured data, especially trends between sampling intervals, and this emphasizes the importance of archived soil samples in long-term experiments.