Index is one of the core components of content based similarity search and the data partition is the key factor affecting the performance of index. This paper proposes a new data partition strategy—key dimension based partition strategy on the basis of the traditional distance based partition strategy, and the index technique accordingly. The key dimension based data partition eliminates the overlaps between twin nodes, and the filtering between twin nodes by key dimension enhances the filtering ability of index. The data partition strategy and index technique proposed can greatly improve the filtering ability of index. Experimental results show that key dimension can be used to improve the performance of index, which is of great significance for accelerating the content based similarity search.