The Childhood Autism Rating Scale (CARS) is an instrument for screening and diagnosis of autism. The present study was performed to assess the interrater reliability of a Swedish version of the CARS when used in a clinical setting. The procedure used mimicked a frequent form of consultation in neuropsychiatry and pediatric neurology. During a restricted time period, both an interview with the parents and observation of the child take place. Often this assessment is an important screening procedure and directs further investigation. CARS was used for rating autistic behavior by two investigators in 25 children. A variant of the weighted kappa statistic (correcting for chance and for degrees of disagreement) showed values between .53 and .75 (indicating fair to excellent agreement). Aspects of validity and reliability are discussed.