Aim: (a) To study the stability of poor language development assessed by a new screening instrument based on parents' recognition of words their 18-mo-old children use; (b) to evaluate the predictability of the less-than-eight-words cut-off used by Child Health Centres (CHCs) in Sweden at the present 18-mo check-up--an assessment based on parents' recall of their children's vocabulary. Methods: All failures of the less-than-eight-words criterion, corresponding to about 10% of 1000 children from a community-based study, were followed up within half a year after the first assessment. The follow-up instrument was an extensive, internationally well-known and structured parent questionnaire, standardized for Swedish children. Results: Almost half of the followed-ups performed below the 10th percentile and more than two-thirds were found below the 20th percentile. Thus, after a few months about one-third of the children had caught up to a fairly normal level of performance. Two of the children who had recovered performed above median. More relaxed cut-off criteria were applied ad hoc , reducing the proportion of children with a fast recovery.