We used the so-called deterministic annealing algorithm due to Rose and Gurewitz by the classification of patent documents. A C++ program based on this algorithm was run first on some artificially constructed data and the results were good. Then we tested it on data sets obtained from some already classified patents. The conclusion we reached is that this algorithm provides an alternative classification to the one used in the US Patent Classification System. Key Words. Unsupervised Classification, Deterministic Annealing, Data Mining, Patent Classification