Risk assessment is an important tool for informing many kinds of societal decisions. When adverse outcomes (e.g., automobile injuries) are observable directly, risk assessment is usually straightforward and involves few questionable assumptions. However, in most cases risk assessment is a complex process because risk information cannot be measured directly. This necessitates the use of complex modeling involving questionable assumptions. Decisions on cleanup of environmental contamination, decisions to site permanent facilities for the disposal of radioactive waste, and decisions on securing nuclear facilities and materials against terrorist threats require sophisticated risk-assessment calculations. For these applications and others, there is limited experience on which to base estimates of the likelihood and consequences of certain events. The usefulness of risk assessment for decision making is limited by the extent of uncertainty in the analysis. The behavior of complex systems can be difficult to predict because of an imperfect understanding of system parameters (conceptual model uncertainty) or incomplete information about important system properties. Understanding the uncertainties and limitations of risk assessment and conveying those limitations to decision makers in an effective manner remain key challenges for the technical and policy communities.