This paper compares individual and household predictors of underweight among young children in sub-Saharan Africa and India, while also assessing the impact of clustering of weight for age z-scores at the household, community and regional levels. Multilevel statistical models are employed to compare the strength of the correlates of underweight (using weight-for-age z-scores) in six sub-Saharan African countries and four Indian states. The multilevel approach controls for correlation among children resulting from clustering within families, communities, or regions and in addition enables tests for differences in the regional, community and household effects for children from families of different socio-economic characteristics. Findings demonstrate the importance of individual and household level predictors such as age, the size of child at birth, prolonged breast-feeding, recent diarrhoea episodes, and maternal education as predictors of low weight-for-age z-scores across regions. Strong family effects are observed as well as significant community and regional random effects on variation in weight for age z-scores. In some regions, socio-economic characteristics of the household result in significant differences in the household or community level variance in weight for age z-scores, suggesting that the impact of the geographical context varies by socio-economic status of the household.