Readers are idiosyncratic. Recent work with library data has underscored this fact more than ever, but it has also drawn attention to specific patterns – many unexpected – in reading behavior. But what is the relation between borrowing patterns and borrower idiosyncrasy at the broader level of overall library use? Identifying thematic continuities in the books that particular borrowers read or demographic continuities in the borrowers that read particular books is instructive. In order to understand the scope of these patterns as reading practices, though, we should also identify whether they can be distinguished quantitatively from other borrower activity and checkout trends. To do so, this paper explores several approaches to using logistic regression with data from What Middletown Read as a form of predictive modeling. I test the degree to which patrons’ borrowing histories are predictive of whether they borrowed any given individual book. In evaluating predictability, however, I also evaluate the assumptions of modeling predictability itself given the peculiarities of library checkout data, characterized as it is by wide disparities in patron use and book popularity as well as the limitation of physical copies. In light of the latter, I consider in particular the status of false positives, which can be understood as recommendations and thus shed light on patterns of taste as they extend beyond the activities of individual readers.