The failure probability of a passive decay heat removal (DHR) system after a loss-of-coolant accident (LOCA) is evaluated as part of a risk-informed design process for a helium-cooled fast reactor. The system was modeled using RELAP5-3D. The epistemic uncertainties in input parameters as well as the epistemic model uncertainties in the code were assessed and propagated through the model using Latin hypercube sampling. The changes in the design that we investigated reduced the overall failure probability of the system by reducing the impact of the major contributor to the failure probability. Sensitivity analyses led to two unexpected results. First, the key factors affecting the system failure probability are the location of the thermal insulation (inside or outside the hot leg) and the uncertainty in the insulation thermal conductivity. Second, the heat transfer coefficient in the core is not as important as one might expect. Our results show that the heat transfer coefficient in the containment structures is more important. Different methods for sensitivity analysis were applied and gave consistent results. The calculated conditional (given a LOCA) failure probability of the passive DHR system was deemed to be unacceptable and led the Massachusetts Institute of Technology design team to adopt an active DHR system as the main mode of DHR for the gas-cooled fast reactor