Purpose: To evaluate the effectiveness of a screening instrument (the Swedish Communication Screening at 18 months of age; SCS18), derived from the Swedish MacArthur-Bates Communicative Development Inventory, in identification of 18-month-old children who will be severely language disabled by 3 years of age, the authors (a) analyzed which SCS18's component yielded the best prediction, (b) compared the productivity figures of the SCS18 with those of the traditional method of identification, and (c) tried different cutoff criteria of the SCS18. Method: Half of the child health care (CHC) centers in a Swedish county were randomly selected to use the SCS18 (e.g., a checklist supporting parents in assessing their child's word production, word comprehension, and communicative gestures). Remaining CHC centers used an informal assessment. Expressive and receptive language was subsequently judged with an observation for 3-year-olds that is routine in the county. An unselected population of 2,080 children participated at 18 months of age and again at 3 years of age. Result: Number of spoken words yielded the best prediction, and SCS18 was superior to the traditional method. A sensitivity of 50%, however, was not enough, and a stricter criterion resulted in too many false positives to be acceptable as routine. Conclusion: Although the SCS18 has strength, the age of 18 months seems to be too early for identification of severe language disability. (Contains 5 tables and 2 figures.)