This paper proposes the multidimensional clustering technique called GKD-tree method which is fully adaptive to both of tuple and query distributions the method is based on the theoretical analysis of the performance of multidimensional clustering Its algorithm and performance are described it is shown that GKD-tree method can largely reduce the average number of page accesses for several distributions its behavior is analyzed and the modification to the performance GDK-tree method was first developed as the physical database organization of the relational database machine GRACE The implementation issues of GKD-Tree method in the database machine are also discussed.