Multimodal interfaces offer to users different ways to express their intentions using the most appropriate means. For instance, to refer to the object involved in the mentioned action, the difficulty of naming decides of the modality (speech or/and gesture) used to access the referents as shown by De Angeli. However analysis of referring gestures remains based on simple principles which constraint users' gestural expression to unnatural production rules. Our experimental study has shown that the variability of natural gestures is governed by the visual context. In this article, we present a contextual modelling of the analysis of spontaneous referring gestures used in multimodal interfaces. So it becomes possible to cope with the designations' diversity and offer a wider freedom of expression for the user.