For the multi-step-prediction of chaotic time series, a modeling method based on maximized the mutual
information(MMI) is proposed. Firstly, maximize information model in phase space with delay time, embedding dimension
and prediction step is established. Then this model can be solved by using generation algorithm(GA), which determines the
better prediction structure of chaotic time series. Finally, simulations of the Mackey-Glass system and monthly sunspot show
that MMI can determine the better prediction structure and improve the prediction accuracy of chaotic time series.