Robust scientific conclusions are too sparse to inform fully most of the choices that physicians must make about tests and treatments. Instead, ad hoc rules of thumb, or "heuristics," must guide them, and many of these are problematic. Physicians extrapolate from the small samples studied by clinical trials to general populations, but they do so inconsistently. Many physicians live by rules that dictate "not treating the numbers," correcting abnormalities slowly, achieving diagnostic certainty, and operating now to avoid "greater" risk in the future. Yet in each case, historical trends or statistical realities suggest either doing the opposite or investing in more discriminating heuristics. The heuristics of medicine should be discussed, criticized, refined, and then taught. More uniform use of explicit and better heuristics could lead to less practice variation and more efficient medical care.