The objective of this study was to evaluate stream flow predication capability of three hydrological models (PED-W, HBV-IHMS and HEC-HMS) in range of watershed sizes in the Upper Blue Nile Basin. The selected watersheds were Anjeni (1.13 km2), Gumara (1280.73 km2) and Main Belles (3431km2) in the upper Blue Nile basin. Calibration and validation was carried out using the stream flow data for Anjeni watershed from 19861994 and 1995-1998, for Gumara watershed from 2000-2009 and 2010-2014 and for Main Belles watershed from1994-2003 and 2004-2008 respectively. The model efficiency on daily time scale during calibration period for PED-W (NSE=0.76, 0.81 and 0.57), HBV-IHMS (NSE=0.68, 0.79 and 0.59) and HEC-HMS (NSE=0.63, 0.68 and 0.48) were obtained for Anjeni, Gumara and Main Belles watersheds respectively. Similarly for validation period PED-W (NSE=0.6, 0.73 and 0.37), HBV-IHMS (NSE=0.56, 0.79 and 0.55) and HECHMS (NSE=0.52, 0.74 and 0.37) were obtained for Anjeni, Gumara and Main Belles watersheds respectively. Performances on month time step also were varying and better than the daily time scale. The study indicated that the difference in model behavior depended on runoff mechanism. In PED-W, saturation excess is the main direct runoff process. HBV-IHMS model runoff simulation depended on soil moisture status and evapotranspiration, and hence was able to simulate saturation excess flow but not the extent of the saturated area. In contrary, HEC-HMS model simulated stream flow where infiltration excess was the main runoff mechanism. The overall model performance indicated that PED-W model was the most appropriate model to predicate stream flow followed by HBV-IHMS and HEC-HMS model. The result indicates that the models in the highlands of Ethiopia are dominantly dependent on the runoff mechanism dominantly on saturation excess runoff mechanism. Hence, there should be an approach to integrate climate region specific model in our water resource development system for predicting stream flow for ungagged catchments.