In this paper we present an incremental elicitation method to determine the importance of the coalitions of criteria in a multicri-teria sorting method. The method is designed to assign alternatives to predefined categories by comparing their performance vector to reference profiles. These comparisons lead to binary preference indices that are aggregated to determine the membership of the alternatives to predefined categories. We present an active learning process to determine the weighting coefficients modeling the importance of criteria in the aggregation process. Learning examples are generated one by one and presented to the Decision Maker to efficiently reduce the uncertainty attached to criteria weights. The process is stopped when all alternatives can be assigned to a category with the desired guarantee. We present the formal elicitation method as well as numerical tests showing its practical efficiency.