Population attributable risks (PARs) are often used in health sciences because they offer an apparently easy answer to the question as to the proportion of disease cases that could be prevented in a population if one or more risk factors were eliminated. We discuss some problems in the interpretation of PARs that result from the fact that diseases have more than one cause. Moreover, requirements are discussed which have to be met before PARs can give a realistic idea of the proportion of cases of illness that can be avoided.