BACKGROUND: Mild mental retardation is an enduring and impairing condition. Its prevalence has varied widely across different studies from .5 to over 8%, with higher rates in completely ascertained samples. The current study estimates the prevalence of low IQ in the mental retardation range (intellectual disability) in a population sample and examines the factors that relate to educational identification. METHOD: A total of 2,730 children in school years 8 and 9 attending local authority schools were assessed in school with the group-administered Cognitive Abilities Test (CAT). A sample of 304 pupils at high, moderate and low risk of mild mental retardation was selected for in-depth study. This included the individually measured full-scale IQ (WISC-III(UK)), the Wechsler Quicktest of attainments, the Strengths and Difficulties Questionnaire from parents and teachers and an abbreviated version of the Social Communication Questionnaire. RESULTS: Of those selected for the in-depth study, 204 (67%) participated, with a greater proportion from the low risk group. A range of prevalence estimates were calculated using different imputation methods and assumptions about individuals not screened. Rates of pupils with WISC IQ < 70 varied from 5.8% to 10.6%. There were no significant gender differences. In contrast to the high prevalence estimates using the WISC, the proportion of pupils scoring in the lowest stanine on the CAT was as expected. Only 15% of those with IQ < 70 had a statement of special educational needs or attended a school for moderate learning difficulties. Behaviour, particularly social communication problems, predicted educational identification. CONCLUSIONS: The current study produced a high estimate of the prevalence of mild intellectual disability based on the WISC but not on the CAT. The findings highlight that the majority of mild intellectual disability in the UK would not be detected using registers. Cases that are detected by registers are more behaviourally disturbed than others.