In this paper, we propose a semantic model to capture the semantic locality in hypertext access for client-side caching. To characterize hypertext data from the perspective of clients, we define a semantic region as a cluster of semantically related logical documents. A logical document is defined as a sequence of subsequently visited interconnected documents which in turn are composed of a container file and (optionally) a set of component files. This model makes it easy to deal with temporal locality, spatial locality and semantic locality in hypertext access. To verify the proposed model, we use an experimental hypertext system, called HyperDB. We generate a set of workloads and assess the performance of a set of caching algorithms using the synthetic workloads and the experimental hypertext system.