Markov models have been widely utilized for modeling user web navigation behavior. In this paper, we propose a novel adaptive weighting hybrid-order Markov model-HFTMM for Web pre-fetching based on optimizing HTMM (hybrid-order tree-like Markov model). The model can minimize the number of nodes in HTMM and improve the prediction accuracy, which are two significant sources of overhead for web pre-fetching. The experimental results show that HFTMM excels HTMM in better predicting performance with fewer nodes.