Because of the lack or incompleteness of data, a number of maritime risk studies based on fuzzy rule-based approaches have been proposed. However, two drawbacks can be identified in such approaches. The first is the low degree of discrimination of risk results because of the membership function distribution of the linguistic terms describing inputs. The lack of discrimination becomes degenerative in circumstances where the membership functions of linguistic terms are disproportionately distributed. The second is the lack of consideration for different effects of risk results caused by input weights because of the loss of information from the inputs. By developing a new mechanism for constructing rules and fuzzy conclusions, a risk modelling capable of overcoming such difficulties is proposed. The methodology is validated using two case studies selected from the literature. It is concluded that the framework is able to provide better discrimination of risk results. Risk outcomes can be generated with confidence regardless of the membership function shapes of linguistic terms in both the antecedent and consequent parts. It also has the feature that enables risk results to be sensitive to the minor alterations of input weights.