The author presents the challenges and opportunities in the use of the electromyogram (EMG), a signal representing muscle activity, for digital musical instrument applications. The author presents basic mapping paradigms and the place of the EMG in multimodal interaction and describes initial trials in machine learning. It is proposed that nonlinearities in musical instrument response cannot be modelled only by parameter interpolation and require strategies of extrapolation. The author introduces the concepts of intention, effort, and restraint as such strategies, to exploit, as well as confront limitations of, the use of muscle signals in musical performance.