This paper analyses empirically the impact of food aid on production, sales and purchases. We estimate the discrete choice and the level choice using the Ethiopian rural household survey. The panel dimension allows us to deal with food aid selection. Running a panel Tobit with sample selection and endogeneity we find that food aid reduces the probability of being a producer. It increases the one of being a seller and decreases the one of being a buyer only after 2004 that corresponds to changes in the criteria of food aid allocation. Food aid does not affect the level choice.