This paper lies within the domain of supervised discretization methods. The methodology aims at identifying relevant interactions between input and output variables. A new supervised discretization algorithm that takes into account the qualitative ordinal structure of the output variable is proposed. Most existing supervised discretization methods are designed for pattern recognition problems and do not take into account this ordinal structure. A qualitative distance is constructed over the discrete structure of absolute orders of magnitude spaces. The algorithm presented implements a maximization process of this distance. A simple example allows interpretation of the process of choosing landmarks.