Abstract: A data compression system for use in processing diagnostic image data which uses a predictor to predict future data. The actual data is subtracted from the predicted value to obtain data related difference values. The difference values are coded by coding the most prevalent difference by a code other than the least bit code but further coding the most prevalent number by string length coding, with the most common being the coded with a least bit code, and also coding the next most common difference with a least bit code and the subsequent most prevalent differences with sequential least bit codes to provide compression ratios of over 3:1. A procedure for finding the best predictor in noisy data is provided.