This paper examines the effects of uncertainty in a structural health monitoring application. Decision uncertainty acknowledges that classification system decisions are probabilistic. The particular task of interest consists of detecting and localizing which one, if any, of fifteen fasteners is loose in a thermal protection system panel. From laboratory data collected during a three month interval, a benchmark classification system is designed to detect and localize loose fasteners and corresponding accuracies are computed. The performance of this system is measured in terms of probabilities of detection, localization, and false alarm. When the benchmark classifier is applied to an independent test set of over 4,900 trials, the probability of detection is 99.6%, the probability of localization is 98.0% and the probability of false alarm is 1.0%. A method is described for reducing the effects of uncertainty and applied to the benchmark classification system. With this processing, the probability of detection becomes 99.0%, the probability of localization becomes 97.6% and the probability of false alarm becomes 0.3%.