The problem is considered of developing the confidence sets for parameters of the transfer function of a control object (CO) on the basis of observable data. The assumptions as to disturbances (noises), affecting the CO, reduce to a minimum: noises can actually be arbitrary, but instead of them the user must have a chance of adding to the input signal a tentative disturbance independent of them. The procedure of solving this problem is suggested and substantiated, which is developed in the framework of the common diagram "leave-out sign-dominant correlation regions" (LSCR), actively progressed in the modern works of Marko Campi with co-authors, and using the approach, earlier suggested by the author, to the CO reparametrization. The procedure returns the confidence region, which for the finite set of observations with the probability selected by the user contains the parameters of the true transfer function.