Multimodal datasets are datasets that provide multiple complementary streams of data corresponding to different recording modalities. In the context of music performance, a multimodal dataset might include modalities such as audio, musical scores, motion-capture data, physiological data, performer biography data, and so on. Collecting such diverse data can greatly expand the range of research questions we can ask about musical performance, especially those that look outside the details of note onsets and offsets and connect more widely with contextual attributes of the composition, performer, and performance.