Intelligent filtering of multimedia documents such as World Wide Web (WWW) pages is an extremely difficult task to automate. However, the determination of a page's relevance to one's interests is a skill that comes with ease to most humans. This paper outlines a system architecture, developed using Agent Oriented Design (AOD), aimed at providing page filtering within a limited domain. This domain is specified via an explicit taxonomy. A prototype system was built using this architecture, which draws upon a user's innate ability to determine the relevance of web pages to their own information needs. It is argued that the resulting system incorporates the best aspects of existing Automated Collaborative Filtering (ACF) systems, whilst still retaining the benefits of the more traditional, feature based approach. In support of this claim and the general success of these systems, initial evaluation of the system is also reported.