A critical analysis of methods for selecting central events in high-energy proton–nucleus (pA) and nucleus–nucleus (AA) collisions is presented. A sample of event classes in which background fluctuations associated with the dispersion of the impact parameter of each event or the number of participant nucleons are minimal is examined. At the SPS and LHC energies, the numbers of nucleon–nucleon collisions are estimated with the aid of the Monte Carlo event generators HIJING and AMPT, which take into account energy–momentum conservation, and on the basis of a non-Glauber model involving string fusion and a modified Glauber model. The results obtained in this way demonstrate the need for revising the extensively used application of the Glauber model in normalizing multiplicity yields in experimental data on pA and AA collisions in the soft region of the spectrum.