The usefulness of different anthropometric indices to detect nutritional changes at the community level, ie, in a number of children considered as a group, was compared by using data from a longitudinal study from rural Bangladesh which followed up quarterly an average of 413 children aged 6-35 months from December 1984 to December 1987. Weight change, mid-upper arm circumference and weight-for-height responded most quickly to seasonal variations of the food situation. Height-for-age was more responsive to long-term variations. Although similar conclusions were reached when proportions of children below a cut-off point or mean indices were compared, the comparison of mean indices required a smaller sample size to detect changes. The difference in sample size needed ranged from 48 to 61 per cent. All indices varied significantly with age, which suggests that precise knowledge of age is essential for proper interpretation of nutritional surveillance data.