The brain uses sensory feedback to calibrate the performance of complex behaviors. A crucial component of sensorimotor control is the generation and evaluation of behavioral variability. Vocal variability allow the brain to explore the space of available motor commands and compute how best to shape vocal output in order to achieve a particular acoustic target. However, the computational role and physiological basis of vocal variability are poorly understood. Our research uses the songbird vocal control system to examine the sources and functions of variability. Recent work from our lab has demonstrated that songbirds use vocal variability to constrain the speed and extent of vocal learning and that doing so allows them to optimize the dynamics of learning across both the time course of vocal development and in response to a range of sensory error sizes. Furthermore, recordings of single neurons in vocal motor cortex show that acoustic variability is shaped by millisecond-scale differences in the timing of action potentials. Together, these and other results suggest a central role for variability in the shaping of vocal learning and the organization of vocal motor activity in the brain.