This paper consists of the application of the Hierarchical lineal model (multilevel methodology) which takes in consideration the interaction between individual and aggregated variables. It is intented to measure eterminants of student performance in their last year of school in three departments of Tucuman?s province (educational census year 2000). Data has multilevel structure indeed students belong to different schools. The GEE method ?Generalized Estimated Equation?, suitable for conglomerate data, allows to model correlation between students within the same school. So multilevel modeling strategies are more likely to produce unbiased estimators than Least Squares, which suppose independence between observations.