Complex motor skills, such as speech or playing a musical instrument, are not innately programmed, but are learned through a process of trial and error. Learning requires motor exploration and performance evaluation. How are these processes implemented in the brain and what happens in disease? Songbirds provide an experimentally tractable model to address these questions. Like humans, they learn to vocalize, first by listening to the sounds of adults during a sensitive period and then by using auditory feedback to practice and modify their vocalizations. In addition, songbirds possess a discrete cortical–basal ganglia circuit specialized for learning and producing song. Variable burst firing in the cortical outflow nucleus of this circuit, LMAN, drives song variability, and manipulations that abolish burst firing in LMAN eliminate song plasticity. Here, I will describe evidence that neurons in this circuit have access to feedback signals, gradually changing their activity in response to feedback perturbation. In addition, I will show that changing the timing and amount of LMAN bursting is sufficient to drive cumulative changes in the acoustic features, timing, and sequence of song. Subsequent inactivation of LMAN did not restore song, indicating that changes in vocal output were encoded in the motor circuit. Together, these findings highlight the importance of temporally precise burst firing in cortical–basal circuits for motor performance, plasticity, and pathology.