Lee [1983> assembled more than two thousand paleomagnetic directional data from lava flows in 65 sampling sites with ages spanning the last 5 million years. Constable [1992> recently suggested that the virtual geomagnetic poles (VGP) derived from this data base have been preferentially located within two antipodal bands of longitude. We have reanalyzed and updated the Lee data base, yielding a total of 3179 data from 86 distinct sites. Neither the total data set, nor various subsets of the data unambiguously show any large and clear maxima in the longitude distribution of VGPs. The shape of the common site longitude distribution is far more striking with a pronounced (and robust) minimum close to the common site longitude and secondary maxima about 120¿ away from it. The former (first order) feature is equivalent to the far-sided effect discovered by Wilson [1970> and corresponds to a persistent axial quadrupole amounting to about 5% of the axial dipole, as previously found by several authors. The latter (second order) feature could correspond to the VGP biasing effect described by Egbert [1992> although its observed amplitude appears larger than predicted.

Part of this effect (an asymmetry in the maxima) may also correspond to Wilson's [1971> right-handed effect. Given only a small axial quadrupolar component and the very uneven actual site distribution, the expected longitude distribution of VGPs can be calculated: this is found to be in good agreement with the paleomagnetic observations (i.e., rather flat with a moderate maximum near 120¿E). This study emphasizes problems related to the unsatisfactory distribution of sites. The amplitude and significance of other terms, beyond the dominant axial dipole and significant axial quadrupole contribution, remain to be assessed. ¿ American Geophysical Union 1994