A population‐based language‐screening programme started in the western part of Sweden (South Bohuslän) in 1991, with the aim of identifying developmental language disability in 2.5‐y‐old children. The programme was evaluated during 1998 in an experimental study. A group of 105 2.5‐y‐old children was recruited from 24 separate Child Health Centres. A sample of 25 children with a positive screening result was selected. For each of the 25 children, 1–4 controls, i.e. children with a negative screening result, were selected. The results were reclassified blindly by three speech and language therapists and used as a gold standard. Positive and negative predictive values were estimated to be 0.52 (95% confidence interval 0.31–0.72) and 0.96 (95% confidence interval 0.89–0.99), respectively. Baye's theorem was used to calculate the sensitivity (0.69) and the specificity (0.93) of the screening. A method of calculating the sensitivity and specificity of the test method by means of three samples was demonstrated. Conclusion: The data confirm that it is possible to identify reliably children with developmental language disability at 2.5 y of age through the screening programme.