Abductive reasoning can be described as a fundamental step in scientific methodology. Its characterization is nonetheless controversial and subject to deep difficulties. Our proposal consists of an analysis of Magnani’s Select and Test Model in connection with the debate about evidence in medicine and the confrontation between mechanistic and statistic-probabilistic evidence. In relation to the Gabbay and Woods model of abduction, we consider mechanisms as abductive hypotheses, preliminary to empirical trials. As such, mechanistic abductions do not provide empirical evidence. However, evidence is a wide concept that can be understood in many different ways. Indeed, evidence is not always to be understood in terms of correspondence to facts or in terms of empirical corroboration. A mechanistic abduction is sometimes an inferential process that provides some kind of evidence in favour of a causal hypothesis, triggered by the observation of a surprising correlation of facts. Thus, mechanistic hypotheses can be knowledge-enhancing since they serve as a guide for empirical trials.