Abstract: Techniques are disclosed that reduce the computational complexity of PPM-based data compression through use of certain simplifying assumptions that permit faster search for a close-to-optimal PPM model than conventional techniques. The disclosed techniques permit the cost of the computationally-expensive model building task to be amortized over many compression/decompression cycles by maintaining a PersistentModel class, accessible to both the coder and decoder side of the system. This allows the computationally-expensive model building task to be performed only occasionally, as opposed to each time a message is coded. Furthermore, the model-building task is preferably scheduled to run at non-busy times, such that it minimizes user-perceptible service disruptions.