Although people have been repeatedly found to underestimate frequencies of common illnesses and overestimate those of rare illnesses, not much is known about this consistent bias in risk perception, termed “primary bias” in the literature, as well as the origin of its variations among different subpopulations. To fill this gap, we designed and implemented a national survey to compare the perceived and actual prevalence of common illnesses, and to test the hypothesis that numeracy may play an important role in the accuracy of judgments of risk frequencies. Our data were collected through a large online survey conducted in France among a representative sample of the adult population (n = 3,245). The participants were asked to complete a 10-items numeracy scale and to estimate the prevalence of a variety of social conditions and common illnesses such as cancers or heart diseases in the French population by using a percentage scale. The analyses show that (1) participants tend to greatly overestimate the prevalence of conditions affecting small percentages of people, and underestimate those affecting a large percentage of them, (2) the Tversky and Kahneman’s probability weighting function provides an adequate model to represent the discrepancy between the perceived and actual prevalence of these illnesses, and (3) the magnitude of the primary bias varies principally as a function of the respondents’ numeracy. These results suggest that the primary bias that affects perceptions of prevalence of chronic diseases is not fundamentally different from those characterizing other types of probabilistic judgments investigated in the field of psychological and behavioral sciences. They also confirm that numeracy plays a considerable role in people’s ability to transform epidemiological observations from their social environment into more accurate estimates.