The main source of data on birth weight in developing countries is statistics from health facilities, although most developing countries do not produce annual estimates of the incidence of low birth weight from these data. Such estimates would be subject to selection bias as the data are usually limited to babies born within health facilities, and therefore are representative of a subgroup that is markedly different from the overall population of neonates. Since 1990 the Demographic and Health Surveys programme has included questions on recalled birth weight and relative size at birth in 15 national surveys. In this article, we show that these cross-sectional surveys can provide a useful data source for making national estimates of mean birth weight and the incidence of low birth weight. The extent of misclassification of birth weight is, however, too large to use the data on relative size at birth as an indicator of low birth weight at the individual level.