The study of the geomagnetic secular variation in prearcheological times is based upon investigations of the angular dispersion of paleomagnetic results and its variation with latitude. The westward drift of the geomagnetic field observed in historic times and apparently confirmed by archeomagnetic data back 1000 years or more suggests the way in which the present geomagnetic field may be analyzed for comparison with paleomagnetic results. Generalized models of paleosecular variation suppose that the angular dispersion arises from two contributions, one due to variations in the dipole field (dipole wobble) and one due to variations in the nondipole field. Attempts at distinguishing between these two contributions are inherently nonunique. To overcome this nonuniqueness, it has been proposed previously that the present low contribution made by the nondipole field to the total field in the Pacific region has persisted for the past 0.7 m.y. The low angular dispersions measured on Hawaii are then a direct measure of dipole wobble because of the so-called Pacific dipole window. However, recent evidence suggests that the data from lava flows for any single locality may be insufficient to measure paleosecular variation adequately. We have analyzed data for 2372 flows distributed over the earth's surface to see if even gross features in the secular variation can be resolved. Between-site angular dispersions, measured with respect to the geographic axis, have been averaged over 15° latitude strips for rocks with ages in the Holocene, Brunhes, and last 5 m.y. The too few data for the Holocene suggest that secular variation during this time has been lower than it has during most of the last 5 m.y. Although data for the Brunhes-aged lavas appear to be inadequate to give a completely reliable estimate of the latitudinal effects of secular variation, data from lava flows formed during the last 5 m.y. do appear sufficient. To overcome the nonuniqueness problem, we have proposed a model for paleosecular variation (model M) that fits both the paleomagnetic data and the variation expected from analysis of the present field and that is also compatible with current theories of the origin of secular variation. This results in an average dipole wobble of 9° over the past 5 m.y. The best fit to the nondipole dispersion then arises from a combination of two mechanisms for the origin of secular variation. The major contribution arises from a nondipole field that originates from some sort of interaction with the main poloidal field, probably akin to Hide's magnetohydrodynamic wave mechanism. The secondary contribution, about two-thirds the magnitude, arises from a fixed field whose intensity is independent of latitude compatible with Bullard's mechanism of fluid eddies interacting with the toroidal field near the core-mantle boundary. The two types of nondipole field are compatible with Yukutake's subdivision into drifting and standing parts, respectively, and provide a possible physical basis for his analysis. During a polarity transition the standing part of the nondipole field should predominate.