Embodiments of the present invention provide a similarity analysis method, an apparatus, and a system. The method includes: acquiring file fingerprint information of a file to be analyzed; sending an analysis request that carries the file fingerprint information to at least two MDSs; selecting at least one group according to an analysis result returned by each MDS, where the analysis result includes a group number and a similarity of at least one group that has the highest similarity with the file fingerprint information and is found by the MDS; and the MDS locally queries a duplicate data block in the selected group. In this way, each MDS needs to query only a file fingerprint information set of a group that the MDS itself is responsible for, which reduces the amount of data retrieval and waiting time of reading, writing, and locking a database file.