Multi-attribute interpretation with discriminant analysis is an efficient tool to interpret 4D seismic data in terms of reservoir changes. Yet, because this method cannot take into account data uncertainties which are in most cases significant, the interpretation may be questionable. The following paper presents an algorithm based on interval arithmetic, which allows to assess the impact of the data errors on the interpretation. An application is proposed on a heavy oil pool produced by steam injection, where a base and two repetitive surveys are available. It includes two parts: first, the quantification of 4D data uncertainties and secondly, their propagation and final impact on the interpretation results.