Reliable estimates of peak flow and flood forecasting are of greatconcerns in water resources management and development. However, the mostconventional methods rainfall runoff modeling and flood frequency analysis for peakflow estimations are not applicable in areas where there are limitations in gettingadequate recorded hydro-meteorological data. This is the common situation encounteredin Ethiopia and similar developing countries. Hence, an alternative method for peak flowestimation has to be developed.This study has identified the existence of similar problem in Abbay basin and hastried to develop empirical methods for estimating peak flows of 5, 20, 50,100 and 200 years return period. The model has been developed based on flood frequency andregression analysis. The result shows General extreme value (GEV), Preson3 (P3), and GeneralParto (GPAR), distributions are the best fit probability distribution for the basin. Peakflow is regressed with some selected catchment physiographic characteristics such asbasin area, longest stream length, channel slope, and basin circularity ratio and very gooddegree of associations have been observed with a coefficient of determinations (R2)ranging from 0.89 to 0.88. The regression equations can be used to predict the magnitudeof peak flows for ungauged streams in Abbay basin, this should be moreapplicable for return period’s less than200 years. The efficiency of the regressionequations is very good, for all recurrence interval coefficient of determination is rangingfrom 88.59% to 89.77%, root mean square error is ranging from 20.21 to 36.97, meanabsolute error is ranging from 11.21to 26.23 and the Nash–Sutcliffe efficiency’s isranges from 0.896 to 0.892.