In natural environments, most events are characterized by stochastic, “noisy” sensory signals. Noisy stimuli can be reliably identified only by accumulating evidence over time. We present studies designed to elucidate how an uncertain stimulus is integrated over time to produce a percept and how that percept is transformed into a decision. Our strategy is to compare human and rat performance to find fundamental underlying principles. From these principles we have constructed a model that accounts for all observed data. In behaving rats, neurons recorded from various regions of neocortex express the computations inherent to the model.