The publication of guidelines calling for less aggressive treatment of jaundice in newborns has been followed by a reappearance of case reports of kernicterus. These case reports illustrate important issues for writers and consumers of practice guidelines. One issue is the particular salience of identified patients with bad outcomes, and their potentially disproportionate influence on decision-makers. A second issue is whether, when good evidence of treatment benefit is lacking, policymakers should recommend what has traditionally been done, recommend less treatment, or not make recommendations at all. Finally, the cases raise the question of whether treatment guidelines should be more conservative than their authors actually believe is necessary, to take into account the likelihood that they will not be closely followed. We believe that case reports can serve as an important early warning system, but policymakers should be aware of their potentially disproportionate influence. In the long run, patients and clinicians will be best served by guidelines that summarize and acknowledge the limitations of existing evidence, that allow a wide range of treatment options when evidence is weak, and that recommend what the guideline authors actually believe should be done. In the short run a period of readjustment may be required, however, as clinicians become accustomed to guidelines written to be followed, rather than bent.