This paper analyses the national key stage 2 test results for 2300 11-year-old pupils in an inner London LEA. A range of concurrent pupil background data was also collected, including whether pupils spoke English as an additional language (EAL), and if so, their stage of fluency in English. EAL pupils at the early stages (1-3) of developing fluency had significantly lower KS2 test scores in all subjects than their monolingual peers. However, EAL pupils who were fully fluent in English achieved significantly higher scores in all KS2 tests than their monolingual peers. The negative association with attainment for the early stages of fluency remained significant after controls for a range of other pupil characteristics, including age, gender, free school meal entitlement, stage of special educational need and ethnic group, although these factors effectively explained the higher attainment of the "fully fluent" group. We conclude that EAL is not itself a good guide to levels of attainment, and a measure of stage of English fluency is necessary to interpret associations with test performance. Alternative measures which focus only on the very early stages of English proficiency, such as the QCA "language in common" steps, are inadequate to assess the impact of bilingualism for all but the very earliest learners of English. Given the uneven distribution of EAL pupils across the country, those schools and local education authorities with high concentrations of pupils in the early stages of learning English are likely to be adversely affected in school achievement and attainment tables. The policy implications for national data collection and for the use of such data are considered.