A neuromotor model of handwritten stroke generation, in which stroke velocities are expressed as a Fourier-style decomposition of oscillatory neural activities, is presented. The neural network architecture consists of an input or stroke-selection layer, an oscillatory layer, and the output layer where stroke velocities are estimated. A separate timing network prepares the network’s initial state, which is crucial for accurate stroke generation. Neurobiological significance of this preparation, and a possible mapping of our architecture onto human motor system is suggested. Interaction between timing network and oscillatory layer closely resembles interaction between Basal Ganglia and Supplementary Motor Area in the brain.