This study applied production function analysis for private peasant holdings crop farms inrnEthiopi a. Four major crop producing regions viz., Tigray, Amhara, Oromia and SNNP werernincluded in the study. Three regression models for production fun ction namely, linear,rnexponential and Cobb Douglas were considered and thoroughly assessed {'or stati stical modelrndiagnostics. The statistical model diagnostics and checking suggested that crop productionrnfunction for each of the regions was found to be appropriately represented by the Cobb-Douglasrnproduction function based on data from the 2007/08 (2000 EC) agricultural sample survey. ThernCobb-Douglas production function was first fitted for each region using ordinary least squaresrn(OLS) regression. As expected, the parameter estimates using OLS were mi sleading due to thernoccurrence of several outlying cases and hence robust regression was taken as a viablernalternative. Based on the results of robust regression, many of the parameter est imates took onrnthe expected signs, the R2 values were substantially increased and the standard errors ofrnparameter estimates were decreased at large.rnIn general, robust regression results indicated that farm size, fe rtili zer, seed, oxen power andrnhuman labor were playing a pivotal role for the maximization of crop yield in each of the studiedrnregions. From among these variables, the great contribution was found to be due to farm size inrneach of the regions with SNNP an exception in whi ch the great share was due to human labor.rnHowever, the contribution of education variable for crop yie ld was found to be stati sti callyrninsignificant and received negative sign in Tigray and Amhara regions. Counter to expectations,rnthe coefficient estimate for irrigation variable in Tigray, Amhara and SNNP regions had come tornobtain negative sign though it was not found to be stati stically significant. The product ionrnelasticities for each of the inputs at each region except farm size in Tigray, Amhara and Oromiarnsuggested that the relation between inputs and output was inelastic, i.e., ho lding other {actorsrnconstant, the marginal return to each factor wi ll decrease as more of the factors are used.rnAdditionally, crop production functions revealed that returns to scale were estim ated to be greaterrnthan unity in each of the regions indi cating increasing returns to scale.