This is the second part of a two-part article describing the results of a sensitivity analysis of the Water Erosion Prediction Project (WEPP) watershed model for rangeland applications. The first article presented the sensitivity of the hillslope component due to variations in hillslope input parameters as well as the implications of applying Monte Carlo simulation and regression analysis to estimate the sensitivity of a complex, non-linear model. This second article presents the sensitivity of the WEPP channel component in predicting hydrologic and erosion watershed responses due to variations in estimating channel parameters. Both model sensitivity studies were conducted using information from a representative semiarid rangeland watershed in the southwestern USA. Three major sources of errors were identified in predicting the stream flow discharge at the watershed outlet: 1) errors in rainfall characteristics; 2) errors in estimating the hydraulic conductivities of both hillslope and channel; and 3) errors in representing the antecedent watershed soil moisture condition. Model sensitivity to channel hydraulic conductivity confirmed the importance of channel transmission losses in predicting the flow in ephemeral channels. Estimates of sediment detachment occurring in the channel bed and total sediment yield at the watershed outlet were highly sensitive to total Mannings n, but only slightly sensitive to erodibility and critical shear stress channel parameters.