There is an increasing interest in value added school feedback in many countries. Systems are developed to provide schools feedback about how well their students perform. Where possible school feedback takes into account students' background characteristics, previous achievement scores or ability and provides information about how well schools perform in terms of the average achievement score of their students compared to the average achievement score of other schools with a similar student body.In Flanders, Belgium, a school feedback project has been launched to provide schools in the future feedback on their students' performance at the school or other group level. This feedback will be based on more than one assessment of students' performance and focuses on students' learning gain between two or more assessment occasions.In order to establish a school performance feedback system, tests need to be developed and a reference group to be tested. Next, a (statistical) model needs to be developed to derive value added scores for schools with regard to the learning gain of their students.To derive conclusions about the learning gain, repeated measures are needed. Repeated measures can be analysed in various ways. In general, multilevel models are adopted because of the multilevel nature of longitudinal data. In this paper we compare different multilevel models (multivariate multilevel and growth curve models) to find out which model is most suitable for school feedback on learning gain. Although growth curve models are more popular than multivariate multilevel models, recently a call has been made for making more considerate choices (De Fraine, Van Landeghem, Van Damme & Onghena, 2005).Data of a representative Flemish sample of first grade primary school students in 121 schools are used. Math and reading were assessed three times since the first graders entered primary school: (1) at the start of the first grade; (2) at the end of the first grade; (3) and at the end of the second grade. The following models will be compared: (1) a linear growth curve model with time as a continuous variable; (2) a linear growth curve model with time as a dummy variable; (3) a piecewise linear model; and (4) a multivariate multilevel model. The models are compared in terms of their fit to the data and the information and the meaning of the value added scores that can be derived from. The multivariate multilevel model has the best fit, but provides less information about how well schools perform over time compared to other schools with a similar student body. The piecewise model has the second best fit and provides value added scores with regard to learning gain for different periods. In this study, it is the most appropriate model. However, when measurements increase, it is necessary to check whether a polynomial growth curve model or a piecewise polynomial model is preferable. De Fraine, B., Van Landeghem, G., Van Damme, J. & Onghena, P. (2005). An analysis of well-being in secondary school with multilevel growth curve models and multivariate models. Quality & Quantity, 39, pp. 297-316.It will become a part of an official publication by the Centre of Educational Effectiveness and Evaluation.