The characterization of semiconductor heterostructures from their mobility spectra has originally been proposed, some 12 years ago, as an improvement of Hall effect measurements. However, the application is far from being trivial anti often leads to the occurrence of extra peaks without any physical interpretation. They are generally referred to as "mirror peaks" and are interpreted in terms of computational artifacts. The present paper comments upon such peaks from a set of data concerning van der Pauw test structures. Our main conclusion is that from an experimental point of view, the transverse component (rho(xy)) of the resistivity tensor is more precisely determined than the longitudinal one (rho(xx)). Thus, we are left with the possibility of artifacts within the experimental procedure. Starting from an adjustment of all available information (rho(xx) and rho(xy)), a preferential adjustment of rho(xy) leads to the disappearance of "mirror peaks". As a matter of example, we report the study of an InP-based bilayered structure and show how the mobility spectrum evidences the highly conducting channel of a HEMT device.