This work investigates decision level fusion by extending the framework of subjective logic to account for hidden observations. Bayes' rule might suggest that decision level fusion is simply calculated as the normalized product of the class likelihoods of the various classifiers. However, this product rule suffers from a veto issue. The problem with the classical Bayes formulation is that it does not account for uncertainties inherent in the likelihoods exclaimed by the classifiers. This paper uses subjective logic as a rigorous framework to incorporate uncertainty. First, a class appearance model is introduced that roughly accounts for the disparity between training and testing conditions. Then, the subjective logic framework is expanded to account for the fact that class appearances are not directly observed. Rather, a classifier only returns the likelihood for the class appearance. Finally, the paper uses simulations to compare the new subjective logic framework to traditional classifier fusion methods in terms of classification performance and the ability to estimate the parameters of the class appearance model. © 2012 IEEE.