Every day, new information, products and services are being offered by providers on the World Wide Web. At the same time, the number of consumers and the diversity of their interests increase. As a result, providers are seeking ways to infer the customers? interests and to adapt their web sites to make the content of interest more easily accessible. Pattern mining is a promising approach in support of this goal. Assuming that past navigation behavior is an indicator of the users? interests, then, the records of this behavior, kept in the form of the web-server logs, can be mined to infer what the users are interested in. On that basis, recommendations can be dynamically generated, to help new web-site visitors find the information of interest faster. In this paper, we discuss our experience with pattern mining for dynamic web-site adaptation. Our particular approach is tailored to "focused" web sites that offer information on a well-defined subject, such as, for example, the web site of an undergraduate course. Visitors of such focused sites exhibit similar types of navigation behavior, corresponding to the services offered by the web site; therefore, page recommendation based on usage-pattern mining can be quite effective.