Under the assumption that one of several given models is the real underlying model of the system, a proper auxiliary signal is defined as an input signal that allows one to select the correct model. It is assumed that there is no knowledge prior to the beginning of the application of the auxiliary signal and that detection is to be done within a specified detection horizon. Under the assumption that the noise energy is bounded, a method for the computation of the minimal energy auxiliary signal is given. The new algorithm extends previous work in that it can handle more than two models and certain types of nonlinearities.