Multimodal interfaces are becoming more common, even in the field of safety critical interactive software, mainly due to the naturalness of the interaction that increases the bandwidth between the user and the system they are interacting with. However, the specificities of multimodal interactive systems make it difficult to gather information from the use of modalities and to extract from this information recommendations for improving the multimodal user interfaces. This chapter aims at presenting how abstract information described in models can be fruitfully exploited to improve the quality of evaluations of multimodal interfaces. The approach presented in this chapter combines model-based verification (based on simulation scenario extraction generated from models) and empirical methods for usability evaluation. Our aim is to try to bring together two separated (and often opposite) issues, such as usability and reliability, into the development of safety critical systems. This approach is illustrated via a Space Ground System of a satellite control room, whose multimodal interaction technique is fully described by the means of formal models.