This investigation compares overt judgments about tenable hypotheses to choices in a concept identification task, as a function of stimulus similarity on successive trials. Two mathematical models are tested: (a) A 1-element local consistency version of Restle's concept identification model and (b) the same model with two additional passive states in which hypothesis testing does not occur. Both models successfully predict a decline in percent correct choices from Trial 1 to Trial 2 in one group. The only notable difference between the predictive characteristics of the two models is that only the former model has a tendency to predict zero occurrences of certain response sequences which do actually appear. Three hypothesis judgment strategies were investigated, Cumulative Deductive strategies being dominant early in training and Cumulative Concrete strategies being dominant midway in training. A finding by Berger that the most frequent error in processing information is the failure to eliminate a hypothesis when it is contradicted by feedback on the current trial was confirmed. (Author)