Direct nitrous oxide (N₂O) emissions (DNEs) from croplands are required in national inventories of greenhouse gases. The Intergovernmental Panel on Climate Change (IPCC) guidelines provide an approach using direct emission factors (EF_ds) to estimate DNEs, which are constants for large regions. The goal of this paper is to establish empirical models to account for the temporal and spatial variations of EF_ds, which, apart from the nitrogen addition rate, also vary with a range of environmental factors, so as to enhance the accuracy of regional/national DNE estimates. Therefore, the seasonal/annual DNEs (n = 71) from upland croplands, which are the differences in N₂O emissions between fields with and without fertilizer-nitrogen addition, were used to statistically relate DNEs to regulating factors including the fertilizer-nitrogen addition rate (F_n), and environmental (climate and soil) factors. The multivariate stepwise linear regression results showed positive combined effects of F_n and clay fraction (Clay) on DNEs (DNE = 0.005 F_n + 6.71 Clay – 0.807, R² = 0.61, p < 0.001, Eq. (2)). Furthermore, the nonlinear regression of F_n, precipitation (P), and clay fraction was also adopted for prediction (DNE = 10^？5 ×3.07 × F_n × P × Clay + 0.543, R² = 0.50, p < 0.001, Eq. (4)). Validation with an independent dataset (n = 31) suggested that both models were better predictors of DNEs than the IPCC model, which only depends on F_n. These empirical models may provide simple but reliable approaches for compiling regional/national, and even global inventories of DNEs from croplands. However, both models were restricted to a limited sample size. Understandably, more field observations are still required to further validate the global applicability of these simple approaches.