A considerable body of kinematic data supports the proposal that independent visuomotor channels are involved in the control of the transport and grip components of reach and grasp. These channels are seen as having separate perceptual inputs, outputs and internal processing and are thought by some to correspond to independent neuroanatomical pathways. The idea that different groups of muscles and biomechanical structures can be controlled independently is attractive, but this kinematically-inspired hypothesis fails to take into account the complexity of the dynamic relationships and their interactions within the neuromusculoskeletal system. Inertial, viscous, centrifugal, coriolis, gravitational and reflex cross couplings exist between efferent drives to muscles and resulting body movements. Rotation at even a single joint generates a complex set of dynamic reaction forces and requires coordinated activation of many muscles throughout the body to maintain posture and balance. In this theoretical paper we present a new view of independent visuomotor channels in the form of an adaptive neural controller that can compensate for the above interactions and decouple the relationships between efferent drives to muscles and resulting body movements. At the same time, the neural controller renders all the dynamics (linear and nonlinear), other than time delays, of the neuromusculoskeletal system, unobservable in the visuomotor relationships. Using the geometry of nonlinear dynamical systems we show that, providing certain constraints on the structure of time delays within the system are satisfied, there exists a neural controller that can render all the dynamics of the neuromusculoskeletal system (except for time delays) unobservable in the responses. The controller simultaneously decouples all the interactive dynamics so that each of the m independent inputs controls one and only one degree of freedom of the response. This means that each degree of freedom in a multi-joint response can be controlled by an independent component of the visual input, a behaviour that has long been observed in visual tracking experiments. The controller effectively establishes m independent visuomotor channels. However, rather than reflecting separate neuroanatomical pathways, the independent channels result from a neural controller with convergent and divergent connections to compensate for the interactive nonlinear dynamics within the neuromusculoskeletal system. This new view of visuomotor channels has implications for neural control processes involved in the acquisition and adaptability of skilled perceptual-motor behaviour in general, as well as for the design of robotic controllers.