During accidents involving chemicals, experts can be asked to assess the effects generated. These experts provide distance effects using computer modeling and are faced with a major difficulty: little (or no) information available in order to assess the situation.The objective of this thesis is to suggest a methodology able to take into account uncertainties in the input data for the modeling carried out in emergency situations and to return explicitly these uncertainties to the manager of the crisis.A first step was to evaluate, for a given situation generating a toxic or flammable cloud, the dispersion of modeling results. A ranking of the input variables according to their influence on the final result was established. This phase was carried out on the basis of a sensitivity analysis with a specifically developed strategy.A second phase aimed to establish a methodology for estimating distance effects (in crisis situations), which takes into account the level of uncertainty in the input variables. A methodology for the classification of input operational data was carried out. This methodology is based on two criteria: the sensitivity of the model to the input parameter and the uncertainty about its value (imprecision or variability). On this basis, a new way of using these variables was suggested. Finally, several methods aimed to restore explicitly the results of this modeling were suggested.
