Anthropometric data from the 1982/83 Papua New Guinea (PNG) National Nutrition Survey were analysed to identify geographical patterns of child growth and investigate their relation to a wide range of environmental, dietary and socio-economic variables. Standardized growth scores for length-for-age (LAZ), weight-for-age (WAZ) and weight-for-length (WLZ) were calculated based on an internal PNG growth reference. Hierarchical Bayesian spatial models based on conditional autoregressive (CAR) priors were subsequently used to model spatial patterns in scores and their relation to different sets of covariates. The geographical differences were bigger for linear growth than for increases in weight. Growth was most reduced in parts of Milne Bay Province, Madang Province, the Torricelli/Prinz Alexander Ranges, and in the area occupied by Angan people. Socio-economic status was the most important factor determining variation in growth within populations. Differences in diet and, to a lesser extent, the physical environment were the main determining factor of differences among populations. Covariate adjustment accounted for more spatially structured variation in LAZ and WAZ than in WLZ. All variables indicating higher socio-economic status were correlated with better growth, as was a high consumption of imported and local high quality foods such as cereals, legumes, tinned fish/meat or fresh fish. This indicates that nutritional interventions in PNG should aim at promoting the consumption of such high energy and high protein foods as well as strengthening the general economic base of rural populations.