We analyze complex model processes and time series with respect to their predictability. The basic idea is that the detection of local order and of intermediate or long-range correlations is the main chance to make predictions about complex processes. The main methods used here are discretization, Zipf analysis and Shannon’s conditional entropies. The higher order conditional Shannon entropies and local conditional entropies are calculated for model processes (Fibonacci, Feigenbaum) and for time series (Dow Jones). The results are used for the identification of local maxima of predictability.