Accurately predicting sediment delivery has been a long-standing problem in the field of water resource management. Many different watershed equations and models have been developed such as the Universal Soil Loss Equation (USLE), the Geo-spatial interface for the Water Erosion Prediction Program (GeoWEPP) and many more, however, these models have not always been able to reliably predict in-stream sediment loads. In this study, two scales, watershed and site level, are used to understand where sediment transported in-stream is being produced. At the watershed scale, USLE was used to estimate sediment yield and then different factors such as connectivity topographic indices were applied as discount factors in an attempt to improve these estimates. The different parameters were then compared to turbidity to determine the level of accuracy of each method. It was found that USLE is not able to predict in-stream turbidity levels in the study area watersheds in Virginia and North Carolina. An implicit assumption of USLE is that runoff is produced on steeper slopes and that sediment production occurs on these hillslopes. However, it was found that flatter-sloped areas were highly correlated with in-stream turbidity. It was also found that in-channel and site-specific parameters such as bank height/slope and level of confinement at higher flows were more accurate predictors of in-stream sediment levels. Overall, turbidity and in-stream sediment levels are not well predicted by models that employ USLE. The distribution of runoff source areas, and channel/bank properties appear to be good predictors of sediment production at the watershed scale. These results indicate that sediment production and transport, as conceptualized by common models and equations, often associate sediment source areas with geomorphic and hydrologic processes in ways that are not consistent with the results of this study. Our results show that sediment is most likely being sourced from the channels and in stream areas.