The soil loss on the Loess Plateau, especially in hilly loess areas, is among the severest in the world, and it would be of great meaning to study soil erosion and sediment transport as well as sediment delivery ratio in the gully/channel system in the highly erodible region. Though many models have been constructed in the hilly loess region on the Loess Plateau, the models predicted well only for the large runoff-sediment events. On the other hand, although the physical models usually perform better for small events, the data required for the models are so rigorous that it is not easy to apply them for most of the catchments. In our present study, Tuanshangou catchment, a unit-catchment in the hilly loess region on the Loess Plateau, was selected. Using 95 runoff-sediment events during the period 1961-1969 and by analyzing peak discharge (Qz), the maximum sediment concentraion (Cz), mean discharge of single storm (Qp) and runoff depth (H), five models were constructed given the fact that suspended sediment yield is the function of suspended sediment concentration (Cs) and runoff depth (Ms=CsH). Then, through analyzing the predicted errors of the models, two models of Ms=H(109.2ln(Qz)+546.1) and Ms=H(Cz-141.2) had better modeling precision with mean errors less than 19% and 12%, respectively. Both models presented good modeling precision for the events larger than 30 t/km2, and presented good modeling application for the Shuiwanggou catchment (a neighbored catchment by Tuanshangou catchment). Compared to the previous models, the two models had better modeling results, especially for the small events in the study area. Deep analysis found that, influenced by different flow mechanisms, for the events less than 300 t/km2, the results modeled by logarithmic function was better than that by exponential model. However, the results modeled by exponential modeling was better than that by logarithmic modeling for the events larger than 1000 t/km2. The structures of the two models built in this paper are simple, only composing of two variables H and Qz and H and Cz, respectively. In addition, the model can be applied according to the data acquirement when studying sediment yield in the hilly loess region on the Loess Plateau. The simplicity and easy application of the constructed models would be of assistance in predicting sediment yield (especially for small events) and in constructing soil and water conservation measures in Yellow River basin in the future.