The scientific and public health claim that smoking is a cause of lung cancer or cardiovascular diseases dates back to the mid-1960s. Nevertheless smoking is neither a necessary nor a sufficient condition for lung cancer. One of the main indicators for causality is that, at the population level, smoking highly increases the probability of having lung cancer. A probabilistic concept of causation was developed by some philosophers that could have given conceptual support to epidemiological causal analysis and inference. Yet, it appears that the agreement on the causal status of specific risk factors did not necessarily lead to the adoption of a probabilistic concept of causation by epidemiologists. In this paper I propose a historical analysis of the emergence of the risk factor concept in epidemiology with the objective of highlighting how the question of causality arose. Causal inference in epidemiology has been structured by the famous Bradford Hill’s criteria that were developed in the context of the ‘smoking-lung cancer’ controversy in a pragmatic objective and spirit. Even if there were not analysis of the implicit concept of causation presupposed by these criteria, I will show that there are several interpretations of causation behind these criteria which are more or less assumed by epidemiologists. All this leads us to the question of pluralism or monism with regard to the nature of causality in epidemiology and more generally in biomedicine.