Facing global changes, modeling and predicting the dynamics of soil carbon stock of forest ecosystems is vital but challenging work. Yasso07 is considered as one of the most promising models for such a purpose. We aim at examining the prediction accuracy of Yasso07 on soil carbon dynamics over the whole French metropolitan territory at a decennial time scale. We used the dataset from 101 RENECOFOR sites network, which encompass most of the French temperate forests. The data include (i) measured yearly litter quantity from aboveground organs part from 1994 to 2008, and soil carbon stocks twice at an interval of ca. 15 years (early 1990s versus around 2010). Using Yasso07, we simulated the stock changes (t C ha −1 yr −1 ) per site and compared them with the measured ones. We carried out meta-analyses to reveal the variability in litter biochemistry between different tree organs for conifers and broadleaves. We also performed sensitivity analyses to explore Yasso07’s sensitivity to inputs, including litter carbon quality and initial carbon stocks. At the national level, the simulated annual carbon stock changes (ACC, +0.45 ± 0.09 t C ha −1 year −1 , mean ± standard error) stayed in the same order of magnitude with the observed ones (+0.34 ± 0.06 t C ha −1 year −1 ). The correlation between predicted and measured ACC remained weak (R² < 0.1). There was significant overestimation for broadleaves sites and underestimation for conifers sites. Sensitivity analyses showed that the final carbon stock was weakly affected by litter carbon quality, but a strongly affected by simulation length to investigate and initial soil carbon quality. We revealed both interest and challenges of applying Yasso07 for temperate forests, which reflected the whole state-of-the-art of soil carbon modelling due to lacking knowledge or data on soil and litter carbon quality and fine root litter quantity, rendering high uncertainties for model inputs.