Austin Bradford Hill was once a friend to The Lancet, but, as occasionally happens, friends fall out. The great legacy of his association with the journal, however, was Principles of Medical Statistics. As each edition was succeeded by another--the first in 1937, the last in 1991--he seemed to shift his view about the influence of statistical method on clinical practice from one of assured certainty to one of modest advantage. That change paralleled a move away from an emphasis on the importance of internal validity in the randomized trial to one of understanding the inescapably practical significance of generalizability. Writers on medical research have explored notions of external validity in various ways. One view, for example, is to seek a close correlation between the participants in a clinical trial and patients seen in practice. The argument goes that such a correspondence has to be made before any decision can be taken about whether to apply the result of that trial to the clinical setting. Another view, first worked out by the American logician Charles Sanders Peirce, is that one must simply rely on the informed guess, based on a reasonable estimate of the limits of extrapolation. The tensions between and implications of these two different approaches are worked through using the example of coronary stents. A solution is, perhaps, to write explicit rules of interpretation that provide a framework for judging the strength of a claim to applicability. Five questions are posed, which try to lay a foundation for such a framework.