This paper examines the problem of how to transfer money or other forms of assistance to poor households when one observes some characteristics of households, but not their incomes. This and related issues are often referred to as the targeting problem. The paper first sets out the problem formally as one of minimizing a poverty index given a fixed amount of money available for transfers. Assuming that household survey data are available which include accurate income and/or expenditure information, the solution for the problem is formulated as a non-linear mathematical programming exercise. Using household survey data from Cote d'Ivoire, the technique is applied to both urban and rural areas separately. The paper concludes with a general discussion and suggestions for future research.