Abstract : Production planning is an important and complex function in production management. In order to facilitate the determination of production plans, this study is concerned with multi-level decisional structures. Levels of decisions are based on an aggregation of data, a disaggregation of decisions and suitable temporal horizons. In each level, the decisions are obtained by successive refinements with an appropriate dynamic. However, the ignorance of interactions between levels often leads to an unsatisfactory behaviour of the structure. For this reason, this work focuses on the formalization and the analysis of interactions between levels in order to achieve a global coordination of decisions within the structure. The generic concepts of robustness and consistency are introduced and characterized. The method consists to establish analytical conditions, introduced as additional constraints within the mathematical models of each level. A study model for interactions between levels is analysed (linear programming and graph theory) and the required conditions are deduced by a generic approach. This methodology is implemented for a multi-product, multi-stage, multi-resource planning problem. For the considered elementary structure, two levels based on time aggregation are defined. This kind of aggregation, associated with a dynamic disaggregation process offers advantages for reactivity and autonomy. These theoritical results are extended in approches taking into account implementation considerations. In this framework, production backlogs are introduced and data uncertainty is explicitly taken into account.