Over the last 20 years, research developed by 'School Effectiveness Research' (SER) tried to uncover the determinants of schools' success, which cannot be attributed to the social and cognitive inter-individual differences. These researches have provided evidence to support that school performance depends on several school processes, as shared vision and goals of the school, participatory leadership, collaboration and collegiality within the teachers' staff, an orderly learning environment, clear and fair discipline climate, maximum learning time and parental involvement (Scheerens & Bosker, 1997 ; Teddlie & Reynolds, 2000). However, in recent years, school effectiveness research has become increasingly criticized for being a socially decontextualised body of literature (Morley & Rassool, 1999; Thrupp, 2001). Indeed, SER seemed to eschew many of the central questions about the relationships between the students' social origins and their school achievement (Thrupp, 1999). Particularly, the question of the contextual effect of the social class mix of a school's intake on individual academic performance has been poorly assumed by SER. Some recent studies have also been interested in understanding more deeply the relations that link school process and school composition. These studies are mainly qualitative ones (Thrupp, 1999 ; Gewirtz, 1998 ; Van Zanten, 2001 ; Opdenakker & Van Damme, 2001 ; Duru-Bellat, Danner, Le Bastard-Landrier & Piquée, 2004 ; Lupton, 2004). To our knowledge, few studies have explored this relation with quantitative design. The main purpose of the present study is to fill this gap by estimating the part of between school variance that is explained respectively by a net and a joint variance of school composition and school process. Through the use of multi-level modeling and model rotation (see Hanushek & al., 1972; Grisay, 1997), scores in mathematics from European countries participating to 2000 OECD survey (PISA project) were analyzed. The results suggest that the joint effect of school processes and school composition explains a significant part of between schools variance, independently of the net effects of school process and school composition, within countries in which the between schools variance represents more than 20 % of the total variance of the mathematics scores. The implications of these results for policies of accountability and school improvement are discussed.