The sensorimotor transformations for visually guided reaching were originally thought to take place in a series of discrete transitions from one systematic frame of reference to the next with neurons coding location relative to the fixation position (gaze-centered) in occipital and posterior parietal areas, relative to the shoulder in dorsal premotor cortex, and in muscle-or jointbased coordinates in motor output neurons. Recent empirical and theoretical work has suggested that spatial encodings that use a range of idiosyncratic representations may increase computational power and flexibility. We now show that neurons in the parietal reach region use nonuniform and idiosyncratic frames of reference. We also show that these nonsystematic reference frames coexist with a systematic compound gain field that modulates activity proportional to the distance between the eyes and the hand. Thus, systematic and idiosyncratic signals may coexist within individual neurons.gain field | posterior parietal cortex | reference frame transformation W e are currently at a theoretical crossroads regarding how the brain computes motor commands from sensory information. The linear systems engineering tradition taught that neural circuits are assembled to compute particular transfer functions and that intermediate stages represent quantities with straightforward physical interpretations (1). At each stage, neurons encode information in a single frame of reference, which facilitates simple pooling. Each cell has a unique receptive field or preferred direction. This results in a distributed population code where each neuron encodes similar information in a similar manner. The vestibular-ocular reflex provides an example of this approach. Head rotation, measured by the vestibular semicircular canals using a canal-centered frame of reference, was believed to be transformed directly into oculomotor commands by virtue of appropriate synaptic weights on vestibular and oculomotor nuclei neurons (1). Visually guided reaching provides another example. Visual spatial information was originally thought to undergo a series of discrete transformations from a sensory (gaze-centered) frame of reference in the occipital and parietal cortices to hand-centered in the premotor cortex (extrinsic motor coordinates) and finally, to muscle commands in the primary motor cortex (intrinsic motor commands) (2-6). The parietal reach region (PRR) in the posterior parietal cortex (PPC) was seen as a discrete processing stage in which reachrelated spatial information was encoded using a uniform gazecentered reference frame and passed on to dorsal premotor cortex (PMd) (7,8).An alternative design borrows from connectionist principles and the field of artificial intelligence. In a trained neural network, the organizational principles may be obscure and individual nodes may encode information idiosyncratically (9, 10). Flexibility and computational power are increased when individual nodes code diverse, seemingly random permutations of the input. Subsequent processing ...