Passive systems represent an important aspect of long-term future solutions for nuclear power plant technology. An issue still to be resolved is the assessment of their reliability. This entails the identification of the physical parameters which are most relevant to the system response and the development of fast-running models to be embedded in the repetitive calculations required by extensive probabilistic reliability analyses. In this work, a case study concerning a realistic design of an Isolation Condenser is considered. The most relevant parameters are identified by means of different sensitivity analysis techniques. Then, fuzzy models are devised which efficiently perform the required mapping between the system outputs and the identified relevant inputs. The approach is shown to lead to satisfactory results, the models obtained being fast and sufficiently accurate and providing “don't know” responses when the models are called to extrapolate beyond reasonable limits.