The Soil Water Assessment Tool (SWAT) was implemented to determine annual sediment yields and critical source areas of erosion for the Buffalo River Watershed. Model calibrations were performed by comparing simulated streamflow discharge and sediment concentrations against measured values. Monte-Carlo simulations were performed to identify the most sensitive parameters and the “best-fit” parameter ranges. This study especially highlighted the importance of snow parameters, which, previously had not been identified as sensitive for model simulations. The cover (C) and practice (P) values for croplands had to be reduced considerably from default model values to constrain simulated sediment yields within the observed data range. The model did not simulate an ice-scour event which generated a substantial amount of sediment. The average annual sediment yield simulated by SWAT for the Buffalo River watershed (108,593 ha) amounted to 0.8 tons/ha/yr. The Cazenovia Creek subwatershed contributed the largest portion (45%) of the total sediment yield from the Buffalo River watershed. We attribute the higher sediment yields from Cazenovia Creek to the greater proportion of steep slopes in this subwatershed. The accuracy and reliability of SWAT sediment predictions at the small watershed (second order or less) and storm-event scales will depend on the accuracy of input information, especially the resolution of the landuse-landcover (LULC) layer, the number of rainfall stations used in simulations, and the number of internal sites against which the model has been calibrated.