Can non-clinicians spot preschoolers likely to have autism spectrum disorder by observing their everyday peer interaction? We set out to develop a screening tool that capitalizes on peer interaction as a naturalistic "stress test" to identify children more likely than their peers to have autism spectrum disorder. A total of 304 3- to 4-year-olds were observed at school with an 84-item preliminary checklist; data-driven item reduction yielded a 13-item Classroom Observation Scale. The Classroom Observation Scale scores correlated significantly with Autism Diagnostic Observation Schedule--2 scores. To validate the scale, another 322 2- to 4-year-olds were screened using the Classroom Observation Scale. The screen-positive children and randomly selected typically developing peers were assessed for autism spectrum disorder 1.5years later. The Classroom Observation Scale as used by teachers and researchers near preschool onset predicted autism spectrum disorder diagnoses 1.5years later (odds ratios=14.6 and 6.7, respectively). This user-friendly 13-item Classroom Observation Scale enables teachers and healthcare workers with little or no clinical training to identify, with reliable and valid results, preschoolers more likely than their peers to have autism spectrum disorder.