This study examined gender, socio-economic (SES), and ethnic group differences in academic performance (measured at 14 and 16 years) in a sample of 517 British pupils (mean age=16.5 years). White pupils outperformed their Black and Pakistani counterparts and high SES pupils consistently outperformed their low SES counterparts. Results from two Multiple Indicators Multiple Causes (MIMIC) models showed that controlling for IQ variance minimizes these;group differences. The MIMIC models also revealed that Pakistani pupils and girls tend to underperform academically relative to White pupils and boys, respectively, at 14 years, once IQ and SES have been partialed out. These and other, more specific, findings are discussed with reference to predictive test bias, selection and streaming procedures, and implications for educational policy. (C) 2008 Elsevier Inc. All rights reserved.