In two experiments, the strategies used by subjects playing the logical-deduction game, Mastermind, were examined. In the first experiment, subjects showed improvement resulting from the continued use of a particular strategic action, and the data suggested that the subjects learned the strategy from their transactions with the task. In the second experiment, the question of changes in underlying strategic knowledge of Mastermind was examined. The accuracy and complexity of the subjects' deductions and their use of the previously identified strategy were used to generate a model of the cognitive operations involved in Mastermind. Although there were improvements in the accuracy and complexity ofthe subjects' deductions resulting from continued play, these improvements were unrelated to the use of the strategy. Moreover, the likelihood of making accurate and complex deductions was well accounted for by a Markovian model, suggesting that the deployment of the strategy was not driven by any change in the subject's underlying knowledge structures. Rather, the subjects seemed to use the strategy to create Mastermind situations whose interpretation was fairly easy. The implications for previous work on the issue of Mastermind strategies and the development oflogical-deduction strategies are discussed.Thus, Mastermind seems to provide a useful situation for studying problem solving: it retains many of the formal mathematical properties of the Bruner et al. (1956) task, yet it has continued to be a popular commercial item for more than a decade.The game tree for the standard game being intractably large, Laughlin et al. (1982) studied the performance of With this conceptual framework as their base, Laughlin, Lange, and Adamopoulos (1982) studied strategy use in the logical-deduction game, Mastermind. In the standard version of this game, the solver's task is the deduction of a left-to-right ordering of four color names, called the code. To deduce the code, the solver generates a hypothesis of four or fewer color names. Feedback is then supplied that informs the solver about the quality of the hypothesis. In general, feedback indicates the congruence between that particular ordering sequence and the code. Feedback can be black or white. Each unit of black feedback indicates that one of the color names in the justadvanced hypothesis matches a code member in color and location. Each unit of white feedback indicates that one of the color names in the hypothesis matches a code member in color, but not in location. In the standard game, each code consists of four colors drawn with replacement from a pool of six colors, thus affording 1,296 different codes. Each hypothesis that the subject makes, and its associated feedback, is displayed using a set of plastic tokens. A typical hypothesis and its associated feedback are shown below:One of the important legacies of the work of Bruner, Goodnow, and Austin (1956) was the notion that humans are strategic in their acquisition of knowledge. As is commonly known, Bruner et al. identi...