The often observed relatively large amplitudes in the later part of the Pn signal cannot be explained with the traditional interpretation of Pn in 1‐D crust‐mantle models. To determine the cause of these characteristics we have analyzed in some detail the NORESS array records of Pn from a suite of quarry blasts in S.W. Norway. Application of wide‐band frequency‐wavenumber analysis to these records confirms that there is a slowness and azimuth anomaly associated with the dominant part of the wavetrain and that it is confined to a particular frequency range. Moreover, the scattering source of the anomaly is determined to be at Moho depth, which is consistent with the concept of scattering by topographic relief. We demonstrate the viability of this concept by means of numerical experiments, showing that for realistic models of topography, the energy flux of scattered P can dominate the specular flux (i.e., the flux in the direction defined by the ray crossing a plane interface) for incidence angles approaching the critical angle of Pn. Since the effects appear to be systematic, we have the possibility to calibrate the Pn parameters for event location and velocity determination purposes.