Results from two multiple regression models involving HighWire journal subscriptions for 16 participating universities in the United States, United Kingdom and Sweden for 2003 indicate a highly predictive relationship between the number of article downloads and the number of users, meaning that the size of a user population can be estimated by just knowing the total use of a journal. The relationship is consistent over time and across institutions, and appears to be unrelated to the subject, size or popularity of a journal. It is not consistent across publishers, however, suggesting that an 'interface effect' may exist. The development of a Project COUNTER standard to deal with extreme or abnormal journal usage is necessary if we wish to compare the performance of journals across publishers.