In several hypermedia applications it is usual to face certain user impatience with long waiting times for presenting a media content. In other circumstances, even a short delay may impair the user interactivity or the synchronization among media objects, decreasing the presentation quality or generating an unintelligible presentation. In order to lessen these problems, the use of prefetching techniques is being taken into account in multimedia/hypermedia document presentation systems. This paper investigates the requirements for designing and implementing prefetching techniques in hypermedia formatters. Based on these requirements, the paper proposes a generic architecture to be adapted and reused in specific formatter developments. Besides the proposed architecture, we specify interfaces in order to make easier the integration of the prefetching mechanism with other hypermedia presentation system components, allowing the incorporation of prefetching strategies in formatter implementations currently in use.