Data de-duplication schemes reduce the amount of storage necessary to store a data set by dividing the data into segments and storing a segment identifier on a storage medium in place of each data segment. Each unique data segment is stored in a repository, and duplicate data segments are not stored. Methods and apparatus are provided for distributing data segments across multiple repositories in a data storage system, thereby reducing the quantity of data stored at a particular repository. Segments are assigned to repositories based upon a characteristic of the segments. The characteristic may be the length of the segment or some other value produced by a repeatable, uniformly-distributed function of the segment. The characteristic may be stored on the storage medium along with the segment identifier. The original data may be regenerated by retrieving the segment identifiers and characteristics from the storage medium and retrieving each segment from the repository identified by the characteristic.