By means of a sensing system we are able to capture bowing and ¯ngering actions executed by a violinist during real performances. The aim of this research is to model the relation between those actions and the sound produced. We describe the process for training and optimizing the model by means of neural networks. Given a set of control actions, the model is able to predict the spectral envelopes of the harmonic and noisy components of the sound which are used for sound synthesis. The model is validated by comparing a real recording with the corresponding synthetic sound produced.