The control of movement is highly complex because of the biomechanical redundancy of the musculoskeletal system (Bernstein, 1967). To cope with the large number of degrees of freedom, humans and animals likely rely on a modular control architecture. In other words, the CNS may activate flexible combinations of motor primitives instead of controlling each muscle independently, a motor primitive being a premotor drive generated by some neuronal population (for example, in the spinal cord) that recruits a covarying group of muscles that remain in a fixed relationship during recruitment (Hart and Giszter, 2010). Hence, motor primitives may represent the building blocks of movement organization. An important direction for research is to investigate the neural basis of this organization in the spinal cord, i.e., the neural mechanisms that select the muscle activation patterns required to achieve a behavioral goal.In a recent study published in The Journal of Neuroscience, Hart and Giszter (2010) addressed this question in an innovative way, by developing a new methodology linking the neural input to the motor output. They supported the hypothesis that modularity is directly embodied in the neural circuitry of the spinal cord. The authors performed simultaneous recordings of muscle and neural activities of anesthetized spinal bullfrogs. Using three types of stimulation (muscle palpation, light touch stimulation, and manipulation of limb position), each one involving a different sensorimotor reflex pathway, they recorded EMGs from the bullfrog right hindlimb and neural data from three sites (depths) of the spinal cord.Hart and Giszter (2010) extracted primitives from the EMG data using the linear statistical method of independent component analysis (ICA). Using the standard assumption that the components accounting for the most EMG variance represent estimates of motor primitives (Ivanenko et al., 2005), they proceeded to investigate how the recorded neural activity was related to the extracted components and to the original EMG recordings. If neuronal firing correlated significantly better with component-based than with EMG representations of motor activity, such a result would be a strong argument in favor of the hypothesis that specific neural activities associate with primitives. A novelty of this study was the use of information-theoretic measures, which quantify associations between random variables, including dependencies that cannot be captured by linear regression methods. Hart and Giszter used the mutual information coefficient (MI), measuring the amount of information that the firing rate shared with the independent components or EMG values. This method suggested that neuronal firing was more closely related to independent components than to contraction of individual muscles [Hart and Giszter (2010), their Fig. 4].Hart and Giszter (2010) also showed that the correlation between neuronal firing and motor components occurred for neurons in a specific region of the spinal cord [the intermediate zone (IZ)] (see...