A method is described by which cognitive structures in verbal data can be identified and categorized through numerical analysis and quantitative description. Transcriptions of interviews (in this case, the verbal statements of 40 researchers) are manually coded and subjected to analysis following the AaO (Agent action Object) paradigm. The texts of the interviews are examined first to see whether and to what extent the same words are used in formulating and defining problems, and whether and to what extent the same objects are used. Cluster analysis models are applied, and the ANACONDA (analysis of concepts by means of data-processing) method is employed to initiate an iterative construction of theories and models. A study is made of the grouping of the agents and objects, taking into account how often a particular word occurs. Then both agents and objects are clustered. Logical relations between agents and objects are examined through the AaO paradigm, and objects are transformed into empirically specified concepts by linking objects to respective verbs. Finally, discriminant analysis is used to describe latent patterns of relationships. (WBC)