The algorithm of the Autism Diagnostic Interview-Revised provides criteria for autism versus non-autism according to DSM-IV. Criteria for the broader autism spectrum disorders are needed. This study investigated the validity of seven sets of criteria from the literature, in 1,204 Dutch children (aged 3-18 years) with and without mental retardation. The original criteria (Rutter et al. in "ADI-R Autism Diagnostic Interview Revised. Manual." Western Psychological Services, Los Angeles, 2003) well discriminated ASD from non-ASD in MR. All other criteria (IMGSAC in "Am Soc Hum Genet" 69:570-581 2001; Sung et al. in "Am J Hum Genet" 76: 68-81, 2005; Risi et al. in "J Am Acad Child Adolesc Psychiatry" 45: 1094-1103, 2006) were sensitive at the cost of specificity, bearing the risk of overinclusiveness. In the group without MR, clinicians should decide whether sensitivity or specificity is aimed for, to choose the appropriate criteria. Including the Autism Diagnostic Observation Schedule revised algorithms in the classification, the specificity increases, at the cost of sensitivity. This study adds to a more valid judgment on which criteria to use for specific objectives.