Over the past two decades, a lot of attention by the media, policy makers and researchers has been given to the emergence of a new gender gap. Since the mid-1980s, there has been a shift in the pattern of educational performance on achievement tests and public examinations to girls outperforming boys on almost all areas of the assessed 'cognitive' curriculum (Arnot, David & Weiner, 1999; Gorard, Rees & Salisbury, 2001; Kleinfeld, 1999). Alongside the number of explanations for the 'underachievement' of boys that are being postulated, a lot of strategies are being explored to combat the underachievement of boys (Warrington & Younger, 2001). One of these strategies is the implementation of single- sex classes in co-educational schools. However, the establishment of single-sex classes for core subjects like English and mathematics in co-educational schools in countries like the United States, Australia and England is growing faster than the evidence to support it (Robinson & Smithers, 1999). It is against the background of this inconclusive evidence that we investigate the effects of the advantages single-sex classes and schools have for boys and girls. In this study not only the effectiveness of single-sex versus co-educational classes, but also the effectiveness of single-sex versus co-educational schools on language and mathematics achievements of boys and girls in Flanders (the Dutch speaking part of Belgium) are investigated. Furthermore, we will investigate whether it makes a difference whether single-sex classes are organised in single-sex schools or in co-educational schools. Or in other words, does the gender composition of the school add an effect over and above the gender composition of the class? Data from the LOSO-project (Longitudinal Research in Secondary Education) are used (Van Damme, De Fraine, Van Landeghem, Opdenakker & Onghena, 2002). This project started in 1990 and followed a cohort of more than 6000 students during and after secondary school. Questionnaires were given to parents, to individual students, to teachers and to headmasters. Information about the performance, motivation and attitudes of students, about their social and cultural background, about teachers, about their way of teaching and about the school were gathered. In this study, we focus on the mathematics and Dutch achievement (mother tongue) of pupils at the end of second grade in secondary education (age 13-14). Our dataset for the Dutch achievement consists of 4131 pupils (1973 boys and 2158 girls), 327 classes, 181 Dutch teachers and 53 schools. For mathematics we will work with 4109 pupils (1974 boys and 2135 girls), 179 classes, 329 mathematics' teachers and 54 schools. Because the data are hierarchical structured (students within classes, classes within teachers and teachers within schools), we use multilevel analyses. The results show that after controlling for pupil background characteristics, for language as well as for mathematics achievement, girls' classes obtain the highest scores, followed by co- educational classes and finally boys' classes. At the school level, we found the same results: girls' schools are at the top, followed by co-educational and boys' schools. For boys - but not for girls - the gender composition of the class and the school has an effect on their language but not their mathematics achievement: boys benefit from the presence of girls in classes and in schools. Furthermore, we found that the gender composition of the school is less important than the gender composition of the class: once the type of class (single-sex or co-educational) is taken into account, the gender composition of the schools does not have an additional effect.