A data compressor uses data known to exist on a destination computer for compressing an input data stream by encoding portions of the input data stream as references to matching portions of the known data. The known data is preprocessed to better correlate with the input data stream. The preprocessing includes identifying and modifying internal references in the known data, such as relative offsets and addresses of jump and call instructions in executable code or cross references and hyperlinks in documents. The preprocessing is driven by generating a set of data that describes specific individual modifications or alternatively indicates relationships between the known data and the input data stream, such as code or data block motion, from which individual modifications can be deterministically made. This generated descriptive data is included with the compressed data stream so that identical preprocessing can be applied to the known data at the destination computer based on the generated descriptive data prior to decompression. Preprocessing the known data improves compression of the input data stream since changed internal references that interrupt matches of the input data stream to the known data can be eliminated.