Soil erosion and sediment delivery models have been developed to estimate the inherent complexities of soil erosion, but most models are not specifically modified for forest operation applications. Three erosion models, the Universal Soil Loss Equation for forestry (USLE-Forest), Revised Universal Soil Loss Equation Version 2 (RUSLE2), and Water Erosion Prediction Project (WEPP), were compared to one year of trapped sediment data for 37 forest haul road stream crossings. We assessed model performance from five variations of the three erosion models: USLE-Roadway, USLE-Soil Survey, RUSLE2, WEPP-Default, and WEPP-Modified. Each road approach was categorized into one of four levels of erosion (very low, low, moderate, and high) based on trapped erosion rate data and erosion rates reported in recent peerreviewed literature. Model performance metrics included: (1) summary statistics and nonparametric analysis, (2) linear relationships, (3) percent agreement within erosion categories and tolerable error ranges, and (4) contingency table metrics. Sediment trap data varied from negligible (<0.1) to hundreds of Mg ha-1 year-1. The soil erosion models evaluated could estimate erosion within 5 Mg ha-1year-1 for most approaches having erosion rates less than 11.2 Mg ha-1 year-1, while models estimates varied widely for approaches that eroded at rates above 11.2 Mg ha-1year-1. Kruskal-Wallis nonparametric analyses revealed that only WEPP-Modified estimates were not significantly different from trapped sediment data (p ≥ 0.107). While WEPP-Modified ranked best for most model performance metrics, the time, effort, modeling expertise, and uncertainty associated with model results may discourage the use of WEPP as a forest management tool. WEPP is better suited for researchers and government agencies that have the capability to measure extensive parameter data. Additional sensitivity analysis is needed to expand default parameters for forest roads within the WEPP and USLE models.