Content analysis traditionally implies a limited set of criteria and procedures, comprising what is here called the classical model. Some studies of therapeutic interviews have deviated from 1 or more of these prescriptions, leading to debate about their methodological status. This paper argues that certain common deviations from the classical model constitute other models which reflect responses to problems posed by the data, the variables under study, and investigator personality characteristics. Rationales for 2 such models are presented, studies representing all 3 models are reviewed, and certain issues common to this body of research are discussed.* Bollard and Auld (1959, pp. 10-12) are the only analysts working in the period covered by this paper, to the writer's knowledge, who made a formal distinction between scoring, contextual, and summarizing units. The scoring unit is the entity that is