We report new paleomagnetic data from ∼2.5 Ga dikes of the Superior Province and derive estimates of paleosecular variation. When combined with data from coeval dikes of the Karelia Craton (Russia) and older (∼2.7 Ga) volcanics of the Pilbara Craton, these estimates suggest a latitudinal variation of secular variation similar to that of the last 5 million years, but with potentially smaller quadrupole family contributions. The observations are consistent with some numerical models of the field that suggest a more dipolar field when the Earth had a much smaller solid inner core.
 The timing of formation of Earth's inner core is largely uncertain. Labrosse et al.  considered a range of core-mantle heat flux histories and concluded that in the absence of radioactive elements in the core the inner core most likely formed at ∼1 Ga (and no earlier than 2.5 Ga). However, high-pressure and temperature experiments provide some support for K concentrations up to 250 ppm [Gessmann and Wood, 2002]. These concentrations could allow for an earlier formation age, but only by several hundred million years [Labrosse, 2003].
 The possibility of inner core growth at ∼1 Ga naturally raises questions about the nature of the geomagnetic field before this time. Prior studies have concentrated on the paleointensity of the field. Data derived from mineral separates (feldspars) from ∼2.5 Ga dikes of the Karelia Craton (Russia) suggest a field with a strength and variation similar to that of the last 5 million years [Smirnov et al., 2003]. Other authors have argued for a lower paleointensity based on detailed whole rock studies of coeval dikes of the Superior Province (Canada) [e.g., Macouin et al., 2003]. But it is not yet clear whether this is a robust conclusion because of the small number of independent values and the possibility of geological and experimental alteration of whole rock samples.
 At present there are few constraints on the morphology of the Late Archean-Early Proterozoic field. Because such analyses can be based on directional data alone, they are far less susceptible to alteration effects. Below we report new paleomagnetic results from dikes of the Superior province and use these results to estimate paleosecular variation.
2. Field Sampling
 In the Early Proterozoic, the central Superior Province was intruded by the Matachewan diabase dike swarm, which may have been associated with rifting [Fahrig, 1987] or plume activity [Halls and Bates, 1990]. The dikes have experienced no more than lower greenschist facies hydrous alteration. Some dikes contain clouded feldspars, indicating exsolution of Fe-oxides related to slow cooling and emplacement at depth [Poldervaart and Gilkey, 1954; Halls et al., 1994]. Because of our interest in paleosecular variation, we avoided zones where only clouded feldspars have been reported (such as the Kapuskasing Structural Zone) and concentrated instead on other areas with clear feldspars (as shown on the maps of Halls and Zhang ).
 Accordingly, we sampled ten sites (MC1 to MC10) in Ontario along Highway 101 (hereafter, the HW101 region), in an area approximately coinciding with the central Chapleau (C2) region defined by Halls and Palmer . Ten additional sites (MC11 to 19) were sampled to the northwest, along Highway 631 (hereafter, the HW631 region), in an area corresponding to the Hornepayne (HP) region studied by Bates and Halls .
 Six to nine field-drilled core samples were taken from each site. Orientation was done with a Pomeroy sun compass and a Brunton magnetic compass. The width of the dikes sampled varied between ∼1 m and ∼32 m (auxiliary Table S1).
3. Rock and Paleomagnetic Results
 Paleomagnetic directions were measured at the University of Rochester by detailed (generally >30 steps) thermal demagnetization experiments in an ASC TD-48 oven. Temperature increments were adjusted to best record the component with high unblocking temperatures (50–200°C range, 50°C step; 200–400°C range, 25°C step; 400–490°C range, 15°C step; 490–600°C range, 10°C step). The natural remanent magnetization (NRM) was measured using a 2G DC SQUID magnetometer with a 4-cm access and a high-resolution sensing coil configuration.
 All samples revealed two or three NRM components (Figure 1). One component with a random direction was easily removed by heating to 100–150°C. An intermediate temperature component persisted to 300–550°C. The direction of this component was not consistent between dikes and may indicate a variable overprint acquired during the Proterozoic [e.g., Schutts and Dunlop, 1981]. For the majority of dikes, the characteristic remanent magnetization (ChRM) was isolated within a narrow (10–25°C) temperature range above 550°C (Figure 1). Such behavior suggests a nearly pure magnetite remanence carrier. This inference was borne out in a series of rock magnetic investigations at the University of Rochester.
 Low-field thermomagnetic curves (κ(T)) were measured in argon from −192 to 700°C using an AGICO KLY-4S magnetic susceptibility meter. With the exception of one dike (MC13b), the κ(T) curves reveal the presence of a single magnetic phase with a Curie temperature between 570–580°C (auxiliary Figure F1). The presence of nearly stoichiometric magnetite is further supported by a characteristic peak observed at −153°C, associated with the Verwey transition (auxiliary Figure F1).
 Both normal and reversed directions of ChRM were isolated during thermal demagnetization (auxiliary Table S1). Using previously published data for the C2 [Halls and Palmer, 1990] and HP [Bates and Halls, 1991] regions as a reference, we identified four sites (MC1, 3, 4 and 12; all with positive inclination) characterized by anomalous directions. The data from these four sites differ by >50° from the means of the remaining data so it is possible they record geomagnetic excursions. However, because there are older and younger dikes in the area, we believe it is more likely that these sites are not part of the Matachewan swarm and we therefore exclude them from further consideration. Dike MC13b, which was characterized by a very low NRM intensity and unstable directions (unblocking temperature ∼350°C) was also excluded.
 The accepted directions plot in two distinct groups corresponding to sampling regions (HW101 and HW631) (Figure 2a). Using a quantile-quantile test, we cannot reject the null hypothesis that the declinations and inclinations within the HW101 or HW631 group are Fisherian at the 95% confidence level (Mu = 1.120, Me = 0.496 (HW101) and Mu = 0.785, Me = 0.687 (HW631)) [Fisher et al., 1987].
 An F-test shows that the group means are distinct at the 99% confidence level. The mean declination difference between the two groups, however, is consistent with prior geological models for vertical axis rotations in the sampling area. We use rotation angles reported for the Hornepayne area and the central Chapleau region [Bates and Halls, 1991] to bring our paleomagnetic sites into a common reference frame (auxiliary Table S2).
 After rotation, the combined population of directions is Fisherian (Figure 2b, auxiliary Table S2). The statistical properties of the population are not significantly affected by the choice of the rotation angle. We used a bootstrap method [Efron, 1982] to calculate the mean pole (53.5°E, 48.6°N, A95 = 5.2°) from the distribution of virtual geomagnetic poles (Figure 2b). While the location of our pole is close to the prior paleomagnetic pole of [Bates and Halls, 1990] (auxiliary Table S2), it infers a slightly higher paleolatitude (λ ∼ 13.5°).
 To quantify paleosecular variation recorded by our sites, we calculate the angular dispersion of the rotated directions (s):
where N is the number of individual site directions and δi is the angle between the ith direction and the mean direction. The dispersion values were corrected for within-site dispersion (sw) related to intrinsic variation within a lava flow and experimental uncertainty using:
where sb is the true-field (between-site) dispersion, n is the average number of samples per site (dike) [Doell, 1970]. Comparable values of dispersion in pole space (S) were also included. Confidence intervals on the dispersion values were estimated using a N-1 jackknife [Efron, 1982]. Using equations (1) and (2) we calculated sb = (12.0 ± 2.5)° from our Matachewan directions. A consideration of additional paleomagnetic data from the Matachewan dikes [Bates and Halls, 1991] indicates that this value is robust.
 To gain a more complete picture of Late Archean-Early Proterozoic PSV, we supplement our estimates from the Matachewan dikes with values calculated from paleomagnetic data from the (∼2775–2715 Ma) Fortesque Group of Australia and ∼2.5 Ga mafic dikes from Karelia (Russia). For both areas published information is insufficient for a correction for within-site dispersion, but the relatively large number of samples and site-level statistics reported lead us to conclude that such corrections are very small (<1°).
 To exclude a component of dispersion potentially associated with motion of the Pilbara craton, we use only the data of Strik et al.  corresponding to a relative pole standstill (representing the Nullagine and lower Mount Jope supersequences, Packages 1 to 7). These data yield s = (8.9 ± 1.3)° for 23 paleomagnetic directions. We calculate s = (11.4 ± 1.7)° based on the D-component of the NRM from the Karelia dikes [Mertanen et al., 1999, Table 3].
 Interestingly, the angular dispersion for the three Precambrian datasets in both directional and pole space are in good agreement with the latitudinal dependence of PSV suggested for the last 5 million years by Johnson and Constable  (Figure 3). In particular, a decrease of s and an increase of S with paleolatitude are observed. This suggests an overall similarity between the geomagnetic field in the Late Archean-Early Proterozoic with that characteristic of the modern dynamo. We note, however, our PSV estimate from the Matachewan dikes, and the values we calculate for the Fortesque and Karelian rocks, plot at the lower bound of the latitudinal dependence of sb presented by Johnson and Constable . Because the Matachewan and Karelian results are from dikes, it could be argued that this trend is related to cooling times slower than those for 0–5 Ma lavas. However, the Fortesque group PSV estimates, which are based on extrusive rocks, are also lower.
McFadden et al.  proposed Model G in which the geomagnetic field is composed of independent geodynamo families: a dipole (or antisymmetric) family, which is linear with respect to latitude and a quadrupole (or symmetric) family that is latitude-independent. Although it is unlikely that the two families are completely independent [Hulot and Gallet, 1996; Johnson and Constable, 1996], Model G is nonetheless useful as an approximation. In particular, the model suggests that the equatorial dispersion is dominated by the symmetric geodynamo family. Together our new PSV estimates suggest a low–latitude angular dispersion lower than that for the 0–5 Ma field, hinting at a more dipole-dominated field in Late Archean-Early Proterozoic times.
 Although the initiation of inner core growth is unclear, the generation of a magnetic field by thermal convection alone seems to require an unrealistically high core-mantle boundary heat flow [Gubbins et al., 2003]. Interestingly, a more dominant axial dipole field (similar to that suggested by the PSV estimates presented here) is predicted by some numerical models incorporating a smaller solid inner core [e.g., Roberts and Glatzmaier, 2001].
 We thank Peter Lippert for help in the field, Pavel Dubrouvine for paleomagnetic measurements, Brian McIntyre for SEM analyses, and the reviewers for useful comments. This research was supported by the NSF.