The distributions of virtual geomagnetic pole (VGP) vectors of the paleomagnetic data from eastern Iceland, western Iceland, and the Canary Islands have been analyzed. For all these data sets, the null hypothesis that each data set represents a random sampling of Fisher's distribution can be rejected with 95% confifence. Similarly, the Dimroth-Watson distribution can be rejected for the Canary Island data, but it cannot be rejected (with 90% confidence) for either of the Icelandic data sets. The orientation of the VGP moment ellipsoids and the increase in angular dispersion with increasing latitude is closely duplicated by a simple stochastic vector model in which the geocentric dipole axis wobbles in a Fisherian manner and the nondipole field anomalies are represented by radial dipoles located on the core-mantle interface and strongly biased to the polar regions. Values of the angular dispersion of the dipole wobble within the interval 80?SD?120 result in vector density functions which resemble the observed densities, whereas values outside this range result in very different densities.