This paper explores the use of household expenditure surveys for assessing food insecurity among people in developing countries. The main objective of the paper is to lay out the background information needed for assessing the reliability, validity and practical usefulness of measures of food insecurity obtained from such surveys. From this stand-point, four main strengths of household expenditure surveys are identified. The first is that they are a source of multiple, policy- relevant and valid measures. These are: (1) household food energy deficiency; (2) dietary diversity, a measure of diet quality; and (3) the percent of expenditures on food, a measure of vulnerability to food deprivation. The second strength is that they allow multilevel monitoring and targeting. The measures can be used to calculate within-country, national, regional and developing-world prevalences of food insecurity and to monitor how these change over time. Because the food data are matched with various demographic characteristics of households, they can also be used to identify who the food insecure are. The third strength is that they allow causal analysis for identifying actions to reduce food insecurity -information that is vital to policy-makers and programme designers intending to reduce food insecurity. Finally, given that food insecurity manifests itself at household and individual levels, as the data on expenditures are collected directly from households themselves, they are likely to be more reliable than those derived from data collected at more aggregate levels. The main weaknesses of household expenditure surveys for the purposes of measuring food insecurity are: (1) they are currently not undertaken on a regular basis in all developing countries; (2) data collection and computational costs in terms of time, financial resources and technical skill required are quite high; (3) data are not collected on the access to food by individuals within households; and (4) although reasonably reliable estimates of food insecurity can be obtained, estimates may be biased owing to various systematic, non-sampling errors.