Fuzzy Multidimensional Databases and OLAP (On Line Analytical Processing) are very efficient for data mining ([3,8]). In [9], we have defined an extension of these databases in order to handle imperfect information and flexible multidimensional queries. An architecture is proposed for knowledge discovery. It integrates fuzzy multidimensional databases and learning systems. We investigate here the problem of generating relevant fuzzy summaries from fuzzy multidimensional databases.