Abstract: An approach of imporving the efficiency for dealing with hadoop small files is proposed in this paper,adding a small file processing module (SFPM) to Hadoop,create indexs of two level according to the file names,and use the preload technology to make the indexs into cache in advance,in order to improve the efficiency of file research and access.When merging files,the strategy of giving up the extra memory space is taken so as to avoid to split one file stored in two blocks,which will reduce the time overhead of file access.The experimental results show that this method can reduce the load of NameNode and enhance its work efficiency of read and write files.