A stochastic version of a knowledge space is developed in which knowledge states are considered as possible epochs in a subject's learning history. It specifies how a subject is channeled through and progresses along a "gradation." The model's application to artificial data is described, based on maximum likelihood methods. (TJH)