This paper illustrates how behavioral models which embody a critical language can be used to describe and predict some categories of classroom events. It defines teaching-learning as a set of events which are mediated primarily by a person, the consequence of which is change in behavior of a second person. A language for describing this phenomenon is needed which should denote those events which may properly be called instruction and those which should be labeled in other ways. The use of behavioral models has three advantages: 1) the models are relatively simple; 2) they require the fewest number of assumptions about the phenomena being described; and 3) there is a rich information bank about behavior analyzed in these terms. Three broad categories of theoretical models seem useful: 1) the Skinnerian-type model of instrumental conditioning; 2) the Miller "drive-cue-response-reinforcement" model; and 3) the Bandura social-learning model. The problems in using behavioral models are of two kinds--those which are the consequences of extending a model by analogy to a new set of phenomena, and those which occur when communicating with investigators who do not use these models. The essential phenomenon of education is the acquisition of formal symbolic systems for the analysis and synthesis of experience; therefore the critical language of education must be concerned primarily with describing the methods of acquiring these symbolic systems. (MBM)