Nonparametric (Kernel) regression analysis and micro-data from the Family Budget Survey (FBS) are used in this paper to estimate and to compare the Engel curves for food demand of the urban and the rural households. The empirical results suggest that the Characteristic Substitution Effects (CSEs) are not constant but vary considerably with the total consumption outlay. They also suggest that the Working-Leser hypothesis, according to which shares are linear in logarithmic expenditure, is consistent with the food demand patterns in Greece.