In estimating the real risk to the public from an accident at a nuclear power plant, several quantities are important: the probability and consequence of the accident itself and the risk resulting from any mitigating action taken. The uncertainties of the risk associated with the accident seem to be dominated by the uncertainties of the consequence estimates. The current procedure of using “conservative” assumptions (usually at each stage) in the calculations produces an estimate of the risk that is likely to be much too high (by as much as an order of magnitude or more). In and of themselves, conservative estimates as typically made in the licensing process may in fact contribute additional risk by overestimating source terms and thus overestimating benefits of activities such as evacuation. This process, in turn, leads inadvertently to putting major segments of society at greater risk than is necessary by encouraging decisions that have higher risk. The principal areas of concern focus on the treatment of a number of physical processes. These processes are always operative and can be counted on to limit the consequences of a reactor accident. Sufficient credit is not taken for their ability to reduce the release of radioactivity and confine it relatively close to its source. Estimates of risk will improve in direct proportion to improvements in quantification of these phenomena. Empirical evidence from many sources shows that these processes are indeed operative and very efficient in reducing the release of radioactivity. As a result, the policy decisions based on the source term in the event of a major reactor accident must be reassessed.