This study, which aims to determine the design principles of interaction adaptation in adaptive online learning environments, was based on a mixed and comparative type of relational screening model. Modified online Delphi method was employed in order to develop the principles and determine the relative importance. Ninety experts participated in the study. Arithmetic mean, first quartile deviation, standard deviation and intraclass coefficient of correlation were utilised to determine the levels of agreement of the experts on the principles. Two hundred and seventy studies which included parameters utilised during the design of interaction were analysed by using content analysis and the robustness of the findings was categorically evaluated, and it was found that experts had almost full agreement on all the 10 principles. Furthermore, the agreement on principles between the observers (intraclass correlation coefficient = 0.903) was found perfect. The obtained principles that indicate the common inclinations both in the literature and of the domain experts should be employed.