AbstractNumerical models have been widely used in risk studies to estimate how hypothetical releases of radionuclides will disperse and lead to exposure of the population. In this context large numbers of possible scenarios are treated; absolute accuracy in simulating each particular situation is not necessary as long as the correct statistical distribution of consequences emerges. Since the Chernobyl accident there has been more interest in the use of numerical models as tools to assist in real‐time in emergency procedures if and when a nuclear accident occurs. However, the use of models in a real situation is distrusted by many people, who feel that the only way to assess an emergency situation is by making radiological measurements. This paper addresses uncertainties involved in modelling and argues that a combination of both approaches should be employed, using numerical modelling as an aid in intelligent interpretation of measurements, and suggesting priorities for international exchange of radiological dat