We describe the design and performance of a caching relay for the World Wide Web. We model the distribution of requests for pages from the web and see how this distribution affects the performance of a cache. We use the data gathered from the relay to make some general characterizations about the web.