We can perform a diverse range of innate and learned behaviors – from running and jumping to a swift tennis serve – with remarkable coordination and grace. While skillful execution of innate and learned behaviors is supported by sensorimotor striatum, how this region encodes behavior remains poorly understood. Studies based on learned skills suggest striatal neurons represent skills as neural sequences, with each neuron locked to a feature of the stereotyped motor output. However, whether this coding scheme generalizes to innate behaviors or emerges through learning is not known. One possibility is that striatum functions as a general controller and codes for movement kinematics similarly across innate and learned behaviors. Such a scheme, however, may cause neural interference if kinematically similar movements appear in distinct action types. A degenerate kinematic code, in which similar behaviors are represented differently depending on context, may solve this problem. To arbitrate between these coding schemes, we combined in-vivo electrophysiology with whole-body 3D kinematic recordings in behaving rats expressing both a learned skill and spontaneous natural behaviors. Examining the striatal code for natural behaviors, we identified a subpopulation of neurons that were tuned to specific behaviors, such as rearing and grooming. Activity of these neurons was well-described by whole-body movement kinematics, with ‘trial-by-trial’ activity during a given behavior comparably precise to what has been described for learned skills. Comparing this subpopulation to that encoding movement kinematics in the learned skill revealed an overlap. Intriguingly, we found that similar movements exhibited during innate and learned behaviors were associated with different patterns of activity. This suggests that striatum has a precise but degenerate kinematic code that is a function of context. This many-to-one mapping between activity and behavior may thus afford the motor system the flexibility to refine behaviors in context-specific ways without undue interference across behavioral domains.