We seek to establish whether or not secular variation, the rate at which the magnetic field is changing in time, is a function of field direction: Is secular variation anisotropic? From a large number of paleomagnetic lava data, many of which record polarity transitions, we find that vectors from stratigraphically adjacent flows are most (least) correlated during nontransitional (transitional) periods. Since volcanic activity is unrelated to, and therefore uncorrelated with, magnetic secular variation, this relationship indicates that secular variation is enhanced during transitions, perhaps by a factor of ~2. Thus secular variation is anisotropic, being a function of the deviation of field direction from an axial dipole. Despite this spatial dependence, we find little convincing evidence that transitional fields exhibit persistent recurring quasi-stationary states which manifest themselves in terms of midlatitude virtual geomagnetic poles (VGPs), as has sometimes been hypothesized. On the other hand, we identify tentatively some relative quiescence in secular variation during midtransition when VGPs fall near the equator. Numerous subsets of the database are examined for consistency and statistical errors are estimated by bootstrap analyses. Statistical models of secular variation need to incorporate the information content in serial correlations, like those analyzed here, from stratigraphic sections if the lava data are to be fully exploited. ¿ 2000 American Geophysical Union