We have developed a fast clustering method FlexDice for large high-dimensional data sets[10]. General clustering methods including FlexDice may be able to find data groups consisting of similar data objects, but they have difficult problems of setting some input parameters to suitable values and showing features of clustering results intelligibly. Then, in order to construct a clustering system with user-friendly interface, we propose a feature extraction method for clustering results. We find a feature of clustering results by using FlexDice again and extracting clusters which differ widely from the distribution of data objects in each attribute with ordinary clusters.