Objective: To evaluate two methods for identifying speech and language problems in preschool children. Design: Prospective population based study. Setting: Inner London. Participants and methods: 37 health visitors were randomly assigned to use a structured screening test (18) or a parent led method (19). Of 623 eligible children aged 30-36 months, the parents of 582 agreed to participate (353 using the structured test and 229 the parent led method). Main outcome measures: Children were assessed by a speech and language therapist blinded to the test result, using the Reynell developmental language scales. Children were classified as having “severe language problems” if the Reynell score was below the third centile for receptive language and as “needing therapy” if the Reynell score was below the seventh centile for receptive or expressive language and clinical opinion. Results: Reference assessments and usable scores were obtained for 458 (97%) of the 474 children screened. 98 (21%) children had severe language problems and 131 (29%) needed therapy. The sensitivity and specificity for the structured screening test were 66% (95% confidence interval 53% to 76%) and 89% (85% to 93%) respectively for severe language problems and 54% (43% to 65%) and 90% (85% to 93%) for those needing therapy. The sensitivity and specificity for referral by the parent led method were 56% (40% to 71%) and 85% (78% to 90%) for severe language problems and 58% (44% to 71%) and 90% (83% to 94%) for those needing speech and language therapy. Conclusions: Both approaches failed to detect a substantial proportion of children with severe language problems and led to over-referral for diagnostic assessments. Screening is likely to be an ineffective approach to the management of speech and language problems in preschool children in this population.