Soil erosion models, typically applied at basin and watershed scales, are rarely evaluated at agricultural field scales due to the lack of spatially-distributed time series data. A novel unmanned aerial vehicle (UAV) methodology was used to quantify farm-field soil erosion from nine UAV surveys and structure-from-motion (SfM). Using a semi-distributed approach, we evaluated soil erosion estimates from the Universal Soil Loss Equation (USLE) and Water Erosion Prediction Project (WEPP). The annual erosion rate, measured with the UAV methodology, was 18.83 t ha⁻¹ yr⁻¹, with USLE and WEPP predictions of 26.23 t ha⁻¹ yr⁻¹ and 16.41 t ha⁻¹ yr⁻¹, respectively. Modelled annual and sub-annual erosion rates with WEPP were within the upper-limit of predictive accuracy, while the USLE tended to systematically overestimate soil erosion rates. These outcomes have implications on the efficacy of conservation efforts, which is highlighted through a discussion and comparison of different best-management practice applications.