Today's growing heterogeneity of end user devices makes it crucial for application developers to deploy applications on as many devices as possible with an acceptable effort. Furthermore, new modalities like speech and gestures allow for more natural interaction especially with small mobile devices. It is hard to simply extend existing Web applications to support multimodal interactions. Hence, it is necessary to model modality-specific aspects in applications and their user interface description languages respectively. The Device-Independent MultiModal Mark-up Language (D3ML) addresses these issues and is described in this paper.