The invention concerns a method for automatic description of an unknown multimedia object, which consists in associating the unknown object with several types of reference multimedia objects each time based on a membership probability particular to each type concerned (G), the method including a step which consists in measuring at least one physical characteristic on the unknown object (F) and in comparing it with measurements of characteristics representing best the reference types, the method further comprising the step (H) which consists in using for each type at least one probabilistic relationship providing a membership probability to the type concerned based on the result of the comparison of the characteristics of the unknown object and of the type, the method further comprising the step which consists in using the probabilities of membership to the various types thus obtained in combination with a series of affinity relationships between types, so as to select memberships which both constitute the majority in terms of probability and which designate one another by their affinity relationship, and so as to exclude memberships with less affinity with the selected types.