In the present paper the correlation between a basic (normal) classification of a sample of 100 pupils and certain model classifications is examined, the latter being constructed through aimed deformations of the previous one. Such deformations may, in practice, be brought about e. g. by evaluating different properties of the sample in the course of examinations, by using different methods in the examinations etc. Definite types of deformations are sought which lead to an appreciable decline in the coefficient of correlation. It is shown that the classification must be subjected to relatively heavy disturbances for the correlation to be thrown into a state where it must be tested as actually non-existent. The construction of model classifications proves, in this way, to be a useful heuristic means in the pedagogical research.