Interpreting medical test results demands statistical reasoning on the part of doctors: How great is the probability of falsely diagnosing an illness based on a positive test result? This probability can be determined with the assistance of Bayes's Rule. Several studies show, however, that doctors often experience problems with these kinds of Bayesian inferences. We demonstrate that doctors' judgments can be considerably improved when numerical information is presented in a form easily accessible to the ways humans process information. This is not the case with the utilization of applied probabilities that has become customary, but is the case when the problem is presented in terms of "natural frequencies" that result from the counting of observed isolated cases in a natural environment. In a series of studies, we varied the representation format of the relevant statistical information. If the information was not presented in the form of probabilities or percentages but simply in natural frequencies, medical experts, as well as laymen, were able to improve their judgments significantly. And dealing with probabilities and percentages can also be easily learned: Two training programs, which showed how probabilities can be translated into natural frequencies, placed the participants in the position of being able to obtain very good results when solving these tasks. Finally, we discuss the impact of a comprehensible risk and utility communication on the doctor-patient relationship.