Based on measurements of responses of a distributed population of middle temporal (MT) neurons made in macaques practicing a fine direction discrimination task, we previously reported that behavioral choices in that task had a stronger association with high precision neurons than with those of lower precision. Consistent with this observation, pooling schemes that up weighted the activities of high precision neurons were found necessary to account for psychophysical thresholds (SFN, 2003). We have now tried to understand how both the psychophysical performance and the distributed pattern of neuron-choice association observed in our experiment can be accounted for in a computational model. We used the recorded activities to construct a continuum of models with various pool sizes, different patterns of inter-neuronal correlations, and read-out mechanisms. We used both uniform read-outs that assume diffuse neural circuits unchanged by learning (e.g., Shadlen et al 1996) as well as models that adaptively learn...