Understanding the variations and controls of soil organic carbon (SOC) at different spatial scales can help in selecting edaphic and environmental covariates that enables us to model SOC more accurately. The present study investigated the distribution characteristics and controls of SOC content at various spatial scales, including a deep soil core (204.5 m) taken from land surface down to bedrock (plot scale), two toposequences with different slope aspects (slope scale), and eighty-six soil profiles along a north-south transect under different land uses (regional scale) in China's Loess Plateau. The results showed that SOC content at different spatial scales decreased exponentially with increasing soil depth, but the rate of reduction differed at various spatial scales and in soil layers at different depths. For the deep soil core, the SOC content and the average rate of reduction with depth in the 0–15.5 m soil layer were significantly higher than the corresponding values of the 15.5–34.5 m and 34.5–204.5 m soil layers (p < 0.05). For the toposequences with varying slope aspects, SOC content in the 0–50 cm soil layer declined rapidly with increasing depth; while SOC content in the 50–200 cm soil layer showed relatively no change. There was no significant difference of average SOC content at depths of 0–200 cm for forestland and grassland considering slope aspects that differed or were the same (p > 0.05) due to the similar climatic conditions. However, SOC content within 0–500 cm soil profile under different land uses along the north-south transect exhibited a significant difference (p < 0.05), following the order of farmland (4.94 ± 1.23 g kg−1) > forestland (3.01 ± 1.45 g kg−1) > grassland (2.03 ± 0.68 g kg−1); moreover, the mean SOC content of the 0–500 cm soil profile generally decreased from south to north following the decreasing rainfall and temperature gradient. The average rates of reduction of SOC content in the 0–50 cm soil layer under different land uses (0.0807–0.1756 g kg−1 cm−1) were higher than the values of the 50–200 cm (0.0021–0.0154 g kg−1 cm−1) and 200–500 cm soil layers (0.0001–0.0017 g kg−1 cm−). The SOC content at the plot scale at different depths positively correlated with total nitrogen content. The SOC content at the slope scale was mainly affected by soil water content and saturated hydraulic conductivity, while that at the regional scale was impacted by climate, topography and soil water/clay content. Pedotransfer functions were applied to adequately simulate and predict SOC content at different spatial scales in the studied area, which could provide a foundation to build SOC prediction models and extrapolate the various spatial scales to other loess regions worldwide. Our findings demonstrate the importance of considering the scale effects for efficiently predicting the spatial patterns of SOC and can help in devising better policy to protect or enhance existing SOC stocks.