A "nested" case-control design was used to identify cases from a longitudinally followed cohort of 1236 newborns registered during 1984-1987, living in three socioeconomically different areas. The children had a length <-2SDS (standard deviation scores) at 6, 12, 24 and 60 mo of age using the NCHS reference. The controls were matched for gender, area and month of birth. A logistic regression analysis was used for determining the risk factors for stunting at each age. Postnatal linear growth was also examined in these two groups of children and body size was compared with the NCHS reference and that of upper-middle-class children (n = 240). At 6 mo of age, prematurity and duration of breastfeeding showed a significant association with stunting. At 12 mo, maternal height, birthweight and stunting at 6 mo, while at 24 mo, stunting at 6, 12 and 18 mo were identified as important factors. At 60 mo, no other factors besides previous stunting could be identified. The mean height reached at 60 mo showed a deficit of 6 and 13 cm for the controls and the cases, respectively, compared to the NCHS reference. Twenty-eight percent of the children from the two poor areas who were stunted at 6 mo had improved by 60 mo of age. The risk factors for stunting varied at different ages, relating more to feeding at early ages and to previous stunting, predominantly at higher ages. The linear growth showed that faltering increased with age when cases and controls were treated separately. Recovery from stunting could also be demonstrated.