Historically, theoretical predictions of human auditory perception have not agreed with experimental measurements. We have previously demonstrated that using signal detection theory to analyze the outputs of deterministic computational auditory models yields more accurate predictions of experimental performance than traditional approaches. However, discrepancies remained between predicted and actual performance. In this paper, the effects of stimulus uncertainty and neural variability on the detectability of a tone in noise are studied. The results suggest that remarkably accurate predictions of detection performance can be generated when such uncertainty is incorporated into the problem.