Due to the increasing complexity of industrial systems, it is necessary to demonstrate that all the risks are under control. In relation to this context, some approaches were developed to provide more integrated risk analysis covering globally technical, human, organizational and environmental risks. One of these approaches, developed from several years ago by EDF and called "Integrated Risks Analysis" (IRA) seems to be a relevantsolution by proposing multidisciplinary, generic and unified model. However, these approaches should be improved with regards to modeling and processing of knowledge (mainly provided by statistical experience feedback and expert judgments). Indeed, this knowledge contains different types of uncertainty that influence the relevance of the results provided by these studies. If the probabilistic framework is well adapted to deal with the uncertainties due to the natural variability of a physical phenomenon, it will be better to choose another framework to deal with the uncertainties due to the lack or imperfection of knowledge. One of the main issues is to develop an approach adapted to the modeling and to the treatment of the different types of uncertainty to provide the most relevant results to the decision-makers. To satisfy this objective, the evidence theory is an interesting framework to provide more coherent and more flexible modeling and treatment of knowledge and their uncertainties. Based on this theory, our works propose three main contributions for the modeling and the treatment of the uncertainty in risks analysis. The first one concerns the formalization of the risks quantification by expert judgments by proposing various tools for a more complete and flexible assessment (risks quantification). The second contribution provides an evidential risks model allowing to represent the different knowledge expressions a more coherent way and to estimate all the risks (risks analysis). Finally, in the last contribution, some tools are proposed for decision-making in uncertain environment (risk assessment). The implementation of these contributions on a real industrial application highlights their feasibility as well as their contributions with regard to more conventional approaches. It also allows us to identify new prospects development for the IRA approach.
