One embodiment relates to a computer-implemented method for generating difference data between reference and target files. A difference engine implemented using a computer receives the reference and target files. The difference engine performs a first procedure to generate difference data representing the difference between the reference and target files if the size of the reference file is less than a first threshold size. The difference engine performs a second procedure to generate the difference data if the size of the reference file is less than a second threshold size and greater than the first threshold. The difference engine performing a third procedure to generate difference data if the size of the reference file is greater than the second threshold size. Other embodiments relate to apparatus for generating difference data between reference and target files and for reconstructing the target files from the reference files using the difference data.