This paper reviews and evaluates various kinds of two-dimensional multiplicative autoregressive models used for lossless image compression. Beginning with an overview of the conventional two-dimensional multiplicative autoregressive models, the paper develops the main ideas incorporated in the improved models. Finally, results of a comparative performance evaluation based on sixteen digitized images is presented. It is shown that a combination of parsimonious modeling, predictor stabilization, block-by-block model adaptation, and adaptive contextual entropy coding result in excellent compression performance.