Abstract: Huffman coding (302) is first performed on uncompressed data (402) to obtain Huffman data (406). Additionally, the frequencies of one or more variable-length bit strings are also determined (408). At least one bit string is selected (304) when compression savings provided by the at least one bit string compares favorably with compression savings provided by the Huffman data. The Huffman data (406) is modified to account for the selection of the at least one bit string. Thereafter, compression is performed (306, 308) on the uncompressed data based on token replacements for the at least one bit string and, where necessary, on the modified Huffman data (420). In this manner, performance will be no worse than those techniques that employ Huffman coding. Using the present invention, greater compression efficiencies may be achieved without knowledge of the data being compressed while still preserving random accessibility of the data being compressed.