In a previous study, we reported the results of an early language screening program in which 306 children were screened using a parent-report questionnaire sent through the mail (Klee et al., 1998). A sample of the children screened were given clinical evaluations within a month of screening (n = 64) and again 1 year later (n = 36). Although the screening program correctly identified 91% of 2-year-olds with language delay, it produced a large number of over-referrals. In the present study we examine a revised screening criterion designed to reduce the number of false positives. The revised criterion generated fewer positive screens overall than the original and resulted in improved specificity (96% vs. 87%) and positive predictive value (77% vs. 51%), while maintaining the high sensitivity (91%) and negative predictive value (98%) of the original criterion. We also propose a screening score based on the new criterion, designed to inform the process of deciding which children to bring in for further evaluation.