A closely related type of problem in MCDA to that of sorting is clustering. As in Data Mining, in MCDA too, no unanimous definition of clustering is given. We address this by formally defining several types of clustering in a MCDA context. We base our definitions on any outranking relation from which different preferential situations can be extracted between the alternatives. We define the ideal result with respect to these definitions and develop a method to find them. We present the implementation of this approach in the diviz software and illustrate it using a well-known problem in MCDA.