Historically, theoretical predictions of human auditory perception have not agreed with experimental measurements. We hive previously demonstratcd that using signal dctection theory to analyze the outputs of deterministic computalional auditory models yields morc accurate predictions of expcrimental performance than traditional approaches [I]. Howevcr, discrepancies remained between predicted and aclual pcrlormance. In this paper, the effccts 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 he generated when such uncertainty is incorporated into the problem.