The large amount of test data becomes a serious problem in SOC testing. In this paper, we propose a method to improve the LFSR reseeding based compression scheme. This method rearranges a given set of test data by merging and partitioning test cubes so that they can be decompressed with a fixed-length LFSR. The compression is done by eliminating repeated patterns in consecutive seeds. A singlepolynomial LFSR is used, so that the decompression process is simple and fast. Besides, it does not need an on-chip decoder. The compression method is very efficient, as experimental results show that it reduces 23.6% of stored data and 34.8% of transferred data compared with the previous methods.