The prognosis for patients suffering from heart failure in Sweden differs from county to county, indicating a need for a comparative study in order to uncover the reasons. Conventional single-level analytical methods, however, underestimate the statistical uncertainty in such studies, leading to an inappropriate ranking of regions. Consequently, public opinion and decision-making may be misled. Conventional methods are also unable to disentangle the extent to which differences in prognosis may depend on individual or regional factors. Multilevel analysis, on the other hand, provides a better estimate of statistical uncertainty, and can both identify and quantify the extent to which differences in prognosis depend on either patient-related or regional factors. Using multilevel analysis, we examined the one-year mortality of 38,343 heart failure patients for the period 1992-1995 in every county in Sweden. Regional differences in one-year mortality were found to be very small, the most relevant factors being individual ones. The limited differences distinguishable at the county level may be explained in part by physician density: the more physicians in a county, the better an individual patient's prognosis.