Abstract: Due to most of current early-warning of engineering risk only give a alarm,but cannot forecast, engineering-project risk early-warning based on entropy optimal model was proposed. Firstly, minimum J-divergence entropy was applied to extract the risk early-warning index and to find out the most effective feature by feature extraction. Then the calculating result was classified to judge state of project with a new clustering algorithm-maximum entropy clustering algorithm. Maximum entropy clustering algorithm allocated index vectors to all of the code vectors rather than the nearest code vector with a ratio of possibility. The algorithm is an improved c-means algorithm. Finally, the case was verified to, the results of forecast were classified by the algorithm to estimate the project-s venture. The experiment results show that the improved algorithm can use to predict the project risk quickly and effectively at engineering analyses. The analytical results are basically identical with the actual situation.