Stunting (low length-for-age) is the most widespread manifestation of growth retardation worldwide. Yet, most nutrition programs use weight-for-age for screening of at-risk children. This study tested whether weight-for-age was an effective screening tool in a severely stunted rural Guatemalan population, using data from the INCAP longitudinal supplementation trial (n = 400). Stunting was defined as length-for-age < -2 SD of the National Center for Health Statistics standards at 3 y of age. Sensitivity and specificity analyses and receiver operating characteristics curves were used to compare weight indicators (weight-for-age, weight velocity and weight-for-length) with length (length-for-age and length velocity) and arm and head circumferences measured during early infancy. Length indicators were clearly superior to weight in predicting stunting (Zda test), and velocities were consistently worse than attained growth. Length-for-age at 6 mo had the best performance, followed by length-for-age at 3 mo, and weight-for-age at 6 and at 3 mo. Velocities, weight-for-length and circumferences were all poor predictors of stunting. Using the cutoff of < -1 SD, length-for-age at 3 mo was the best screening indicator for the early detection of growth faltering. Thus, the current use of weight-for-age, which results in large proportions of at-risk children being missed by screening, greatly limits the potential for impact of nutrition interventions.