The purpose of this paper is to analyze the processes of classification task by means of the application of the information processing models by Gregg and Simon to the stochastic process models. The result was as follows; 1) Subjects adopted consistently the strategy which made them select a hypothesis from the subset of hypotheses that are consistent with the last stimulus pattern presented to them regardless of the number of attribute-dimensions. 2) As the number of dimensions increased, the deviations between observations and predictions of any models tended to be great in spite of the stationarity of the series of responses and the statistical independence of the intermediate responses.