With the enormous growth of information on the Web, Internet has become one of the most important information sources. However, limited by the network bandwidth, users always suffer from long time waiting. Web pre-fetching is one of the most popular strategies, which are proposed for reducing the perceived access delay and improving the service quality of Web server (QoS). A semantics-based pre-fetching model is presented in this paper. This model predicts future requests based on latent intention that the user抯 current access path implies in semantics, rather than on the temporal relationships between URL accesses, which overcomes the limitation of previous pre-fetching approaches. The hidden Markov model (HMM) is employed for mining actual intention from access patterns. Experimental results show that the proposed pre-fetching model has better general performance.