This work focuses on decision problems concerning risk reduction in the case of catastrophic events, which are scarsely predictable but associated to severe expected damage. A cryterium is outlined in order to judge the relative credibility of competing models. The knowledge-decision process is dissected in its components. The steps of the procedure from knowledge to final decision are analyzed. The effectiveness of short-term prediction and the choice of the acceptable risk are also discussed. A few case studies, related to earthquakes, landslides and pollution, are exposed.