most of the research on early school leaving addresses this issue either from an individual perspective (risk and protection factors, personal trajectories, etc.) or from an institutional perspective (prevention programmes, care arrangements, policies to combat early school leaving, etc.). Research on school and classroom scarves on dropping out is much more rare (Galand -Hospel, 2015; HUGON, 2010). However, some work carried out in the United States and in Europe by means of multilevel analyses indicates significant variations in the drop-out rate of pupils according to the school they attend (Goldschmidt -Wang, 1999; Lamote, Van Damme, Van Den Noortgate, Speybroeck, BOONEN -Bilde, 2012; Rumberger, 1995). These school heads are associated with both the composition of school audiences and the organisational characteristics of the schools (e.g. school climate, educational practices) (Christle, Jolivette -Nelson, 2007; Lee Aam, 2003). However, this research does not take into account the class level, which is known to be more important than the school level for many socio-emotional variables (Van Landeghem, Van Damme, Opdenakker, De frairie -Onghena, 2002) and typically include few variables relating to teaching practices (De Witte, cabus, Thyssen, Groot, van den Brink, 2013). The aim of this study is to assess the extent of school and class effet on the risk of early school leaving and to examine the contribution of composition effects and teaching practices to these variations. 3646 colleges (average age 15 and a half) from 64 establishments located in urban areas participated in this survey by completing an anonymous questionnaire. The questionnaire covered the familiar structure and functioning, school history, teaching practices, psychosocial problems, relations with peers, school motivation and engagement, school status and the risk of early school leaving (absenteeism, perceived benefits of early school leaving and intention to drop out). Multilevel analyses indicate significant variations in the risk of dropping out depending on the school attended, but even more so according to the class attended by the pupil. Among the factors investigated, a significant part of the variation between classes is explained both by composition effects (average characteristics of the pupil audience) and by teaching practices. These factors also capture a significant proportion of changes between institutions. The relationship between these factors and the risk of dropping out seems to be partly publicised by pupils’ school motivation and engagement. The implications of these results for the prevention of early school leaving will be discussed.