A new multi-dimensional index structure called RSR-tree is proposed, which based on RS-tree. In RSR-tree, index records of a leaf node are split to ensure the sequence ordering of index records in a leaf node, which reduces the addressing cost of I/O operations effectively when reading data files. The entries of a non-leaf node are split to decreases the overlap between the brother nodes, which reduces effectively the time of reading data from data files. Experimental results on different data sets show that compared to RS-tree, RSR-tree has better comprehensive performance, in regard to tree building and querying. The querying performance is increased and extra cost is not produced.