Underreporting of patient safety incidents creates a reservoir of information that is plagued with epidemiological bias. These include systematic biases such as the practice of reporting minor incidents at the expense of more serious ones. This leads to inaccurate rates of errors and an inability to generalize results to whole patient populations. It leaves reporting incidents, in epidemiological terms, comparable to nonrandom samples from an unknown universe of events. These epidemiological problems lead to a situation where priorities are skewed toward what "we know we know." As "we know what we do not know," for example, gaps in knowledge about serious incidents due to low reporting rates, due caution must be applied in making policy based on biased underreporting. Barriers to reporting contribute to low participation rates and further bias information. Lack of feedback and fear of personal consequences are common barriers. Evaluation of reporting systems indicates reports can be used as tools for learning, but it is not yet possible to monitor improvement in patient safety or measurably prove reduction in harm. Mandatory reporting makes sense from an epidemiological point of view, but there are legitimate fears that it could further reduce reporting rates due to fear of reprisal. Underreporting and the associated biases are a significant problem in realizing the epidemiological potential of incident reporting in health care.