How is information about external stimuli or planned motor commands coded and communicated between different brain regions? This question is one of the most fundamental open questions in neuroscience, and addressing it is likely to revolutionize the field of neural-prosthetic devices. In particular, it will significantly improve the quality of life of the many thousands in need of cochlear implants. Furthermore, the creation of effective motor brain-machine interfaces relies heavily on advances in this field. One hypothesis for the neural code is the Winner-Take-All (WTA) readout, in which activity of single cells, rather than large populations of neurons, shapes behavior. It has been suggested that WTA is used in several brain regions, in particular the middle-temporal (MT) cortex, in which the activity of single cells can account for the observed behavior. However, the computational capabilities of WTA have received little theoretical attention. Therefore, it remains unclear to what extent WTA can account for the observed activity in the brain, particularly in the MT region. I propose to conduct extensive theoretical study of WTA. I will quantify the accuracy of WTA and study the effects of the network architecture, empirically observed noise-correlations, inherent neuronal heterogeneity and different coding strategies, all of which have been shown to considerably affect the accuracy of other readout hypotheses. In addition to studying the conventional WTA, I will construct a novel temporal-code generalization of WTA that can take into account the fine temporal structure of neural responses. My preliminary results show that this readout is superior to the conventional WTA in both accuracy and speed of computation. The primary goal of this research is to obtain quantitative estimates of the accuracy of WTA that will enable us to test its feasibility as readout utilized by the central nervous system by comparing it to the psychophysical accuracy.