The evidence from early intervention studies of autism has emphasised the need for early diagnosis. Insight into the early presentation of autism is crucial for early recognition, and routine screening can optimise the possibility for early diagnosis. General population screening was conducted for 2.5-year-old children at child health centres in Gothenburg, Sweden, and the efficacy of the screening instruments in predicting a clinical diagnosis of autism was studied. The tools used for autism screening comprised the Modified Checklist for Autism in Children (M-CHAT) and an observation made by trained nurses of the child's joint attention abilities (JA-OBS). From the new screening procedure a "definitive" suspicion of autism spectrum disorder (ASD) was raised in 64 individuals in the study population of 3999 young children. Fifty-four of these were clinically assessed in detail. Forty-eight children had a confirmed diagnosis of ASD, three had severe language disorder, and three (6%) were classified as having typical development. The "Positive predictive Value" ("PPV") for the combination of M-CHAT and the JA-OBS was 90%. The combination of instruments used showed promise for early detection of autism as a routine in the developmental programme at child health centres. Trained medical staff is a basic requirement and enables earlier detection and the use of screening tools beyond routine population screening regardless of the age at which a suspicion of autism is raised. (Contains 5 figures and 2 tables.)