Knowledge of nitrogen (N) and phosphorus (P) stoichiometry is essential for understanding biogeochemical cycle and ecosystem functioning. However, large-scale patterns in soil stoichiometry are not yet fully understood along environmental gradients nor over the temporal scale. Using a comprehensive data set and artificial neural network approach, we evaluated spatial and temporal patterns in topsoil N and P concentrations and N:P ratio across China's forests. Our results revealed that soil weathering stage, climatic factors (i.e., temperature and precipitation), and forest types jointly explained approximately 34.1% and 30.4% of spatial variations in soil N and P, respectively. By contrast, only precipitation could explain the variation in N:P ratio, with soil N:P ratio exhibiting a trend of increase along the precipitation gradient. The observed spatial patterns in soil N:P ratio were consistent with previous findings derived from plants and microbes, suggesting that variation in precipitation may induce the imbalance of N:P stoichiometry in forest ecosystems. Our results also indicated that topsoil N:P ratios exhibited a significant increase from the 1980s to 2000s. However, the associations of N:P dynamics with a single element largely depended on forest type. In evergreen forests, soil N:P dynamics were caused by increasing N and decreasing P. Conversely, N:P changes in deciduous broadleaf forests were triggered only by soil N accumulation. Overall, these results demonstrated a stoichiometric shift in soil N:P both spatially and temporally, implying that nutrient imbalance between soil N and P may be accelerated under global change scenarios.