Recently, several authors have described the school composition effect as a methodological artefact (Gorard, 2006 ; Harker & Tymms, 2004 ; Nash, 2003), pointing out that the school composition effect can result from two main categories of methodological bias : model underspecification and predictor unreliability. The first purpose of this paper is to discuss those methodological considerations and to test empirically the impact of the model specification on the magnitude of the school composition effect using reading performance within primary school in French-speaking Belgium. The results show that the school composition effect remains significant even after controlling for the pupils' initial performance, sociocultural capital and non-cognitive dispositions, despite the fact that the total and the between-schools variance explained by the school composition vary greatly according to the introduction of these individual parameters. The second objective is to examine the covariance between the school composition and some organizational variables (organizational culture, principal leadership, teacher collaboration and collegiality) and their joint effect on schools' performance. That second set of analysis aims at questioning the conceptual nature of the school composition effect : direct or indirect effect ?The sample consists of 2164 students nested in 52 schools from the French community of Belgium. The sample is representative of the school composition's distribution into the population. Neither the mean nor the variance of the sample differ from the mean and the sample of the schools' population from French-speaking Belgium.Since students are nested within schools, a multi-level analysis is applied (HLM 6.2, Bryk & Raudenbush). First, a null model (without explanatory variables) is fitted to provide estimates of the variance components at each level. Second, an estimation of the impact of the model specification on the magnitude of the school composition effect is computed using two similar sets of multi-level models (one for the academic school composition and one for the sociocultural dimension of the school composition). First, only the composition variable is entered into the model as a level-2 variable. Next, individual variables are added successively to the school composition variable : prior achievement, sociocultural capital and non-cognitive dispositions. For each model, the total and the between-schools variance explained by a net effect of the school composition is computed. Finally, complete models (with composition and organizational processes variables) are analyzed. All explanatory variables are centered around their grand mean, in order to facilitate computation and interpretation.First, the analysis of the empty model reveals that the part of the variance in language performance that is situated at the school level is 26 % percents of the total variance. Second, it seems that the model specification (here the level-1 model) has a strong impact on the magnitude of the school composition, even if the school composition remains significant (p < .001) after controlling for the prior achievement, the sociocultural capital and the non-cognitive dispositions of the pupils.Third, only few organizational variables appear to be associated with the composition variables. What gives accordance to the hypothesis of a net effect (rather than a joint one) of the school composition effect. All the results are discussed in order to establish the validity of the school composition effect and define his conceptual nature. Dumay, X. & Dupriez, V. (in press). Accounting for class effect using TIMMS 2003 8th grade database: net effect of group composition, net effect of class process and joint effect. School Effectiveness and School Improvement.Dumay, X. & Dupriez, V. (2005), Effet établissement : quelles relations entre composition et processus internes ?, Mesure et évaluation en éducation, 28(2), 67-92.Harker, R. & Tymms, P. (2004). The effect of student composition on school outcomes. School Effectiveness and School Improvement, 15(2), 177-199.Nash, R. (2003). Is the School composition effect real ? A discussion with evidence from UK PISA data. School Effectiveness and School Improvement, 14(4), 441-457. Gorard, S. (2006). Is there a school mix effect ? Educational Review, 58(1), 87-94.