Forest soil is a large carbon (C) pool and plays a pivotal role in the global C cycle. The accurate estimation of soil organic carbon (SOC) stocks in forests is the cornerstone of studying the C budget; however, current assessments of forest SOC stocks are highly uncertain. One of the key reasons for this uncertainty is that most previous studies only used a few soil profiles for their estimation, whereas SOC stocks are highly spatially heterogeneous. To accurately evaluate the plot-level SOC stocks of China's forests, we conducted intensive soil sampling (100 soil cores within a plot) in 33 plots across 11 forest sites from south to north China. The average SOC density (SOCD) of these forest sites was 137.4 ± 12.1 Mg C ha^-1(0–100 cm), with significant geographic variations. The highest SOCD (306.8 ± 7.6 Mg C ha^-1) was observed in deciduous needleleaf forest (boreal forest) in northeast China, while the lowest one (64.8 ± 0.9 Mg C ha^-1) was found in subtropical evergreen broadleaf forest in south China. We also showed that the error of the SOCD estimates obtained from the intensive soil sampling was significantly smaller than that of estimates obtained from the traditional sampling method (5.3 ± 1.3% vs. 24.2 ± 5.6%, with a confidence level of 0.95). Our results suggest that intensive sampling can significantly reduce the uncertainty in forest SOC stock estimation by guarding against the effects of spatial heterogeneity, and provide an important methodological reference for accurately evaluating forest SOC stocks and C budgets in other regions.