We present here new paleointensity data from 39 sampling sites collected from the quenched margins of pillow lavas and dikes exposed within the Troodos Ophiolite (∼92 Ma), formed during the Cretaceous Normal Superchron (CNS), a period of approximately 40 million years when the geomagnetic field reversed extremely infrequently if at all. Monte Carlo simulations suggest that a minimum of 25 estimates are necessary for a reasonably robust estimate for the average field strength. Our data suggest a dipole strength equivalent to the present field or nearly twice the post-CNS average. The mean and standard deviation of the dipole moment (81 ± 43 ZAm2; Z = 1021) from the 57 data points compiled here agree remarkably well with those predicted from the long paleointensity record derived from DSDP Site 522. The new data set for the CNS suggests a picture of a strong and stable field during the period of time when it stopped reversing. Moreover, the similarity of the CNS data with the present geomagnetic field suggests that it is presently in a state of unusual polarity stability.
 After decades of effort, much is known about the behavior of the geomagnetic field. The present dipole moment is about 80 ZAm2. (“Z” stands for Zetta (1021).) We know that the frequency of reversals has changed profoundly through time from periods of frequent reversals (as for the last 50 million years) to periods of no reversals for some 40 or more million years (as in the mid-Cretaceous and the Permian). We know that when the field reverses, its strength drops to extremely low values [e.g., Prévot et al., 1985]. We do not know what causes reversals or what inhibits them, although numerical simulations have provided some tantalizing hints [e.g., Glatzmaier et al., 1999]. The observation that reversals are associated with low field strength suggests the possibility that low fields are a precondition for reversal and that a high average field strength would inhibit the reversal process (e.g., Larson and Olson , but see also Loper and McCartney ). Furthermore, the precipitous drop in field strength over historic times has led some to the speculation that the field is headed toward a reversal [e.g., Hulot et al., 2002]. Yet, if the field is at a relatively high value compared to average [e.g., Juarez and Tauxe, 2000], a reversal could be less likely. The strength of the field during periods of no reversals is therefore of general interest. The data, however, are scarce.
 We have modified the PINT database in two ways. We have added the data included in Goguitchaichvili et al. [2002c, 2002a, 2002b, 2003], Riisager et al. , Shi et al. , Tanaka and Kono , Tarduno et al. , Selkin and Tauxe  and, each of the paleofield estimates were converted to virtual axial dipole moments (VADMs) using paleolatitudes derived from a self consistent set of rotations derived from Besse and Courtillot . Most compilations quote dipole moments in terms of virtual dipole moment (VDM) which do not rely on reconstructed paleolatitudes, but use instead a mean inclination value to estimate magnetic paleolatitude. The quality of the mean inclinations are highly variable and in many cases (when using unoriented material) are not available at all. We prefer the more robust and self consistent practice of using a model for global plate reconstruction to estimate paleolatitude and then calculate a VADM. The differences between VDM and VADM are usually minor however.
 From the database thus constructed we have selected data from 1–160 Ma that were obtained from Thellier-Thellier experiments (with pTRM checks; see Coe et al. ). We excluded explicitly transitional data and data for sites with only one specimen. There are two ways of rejecting site means based on their within site scatter. One is to set a value for the standard deviation (σ) in terms of absolute field (e.g., ≤5 μT). The other is to set a value for the percent deviation (100σ/), say 15%. The former is difficult to achieve for high mean values and the latter biasses against low values. We therefore require sites to meet one or the other for inclusion in our compilation. The updated data set (including several corrections of typos) is available in the MagIC database (http://earthref.org).
 The question arises as to how many data points are required to robustly estimate paleofield strength. We address this with a Monte Carlo simulation, by taking random samples of a statistical geomagnetic field model of the type described by Constable and Parker . We choose the TK03.GAD model of Tauxe and Kent  for this simulation. We randomly draw N values of intensity, convert them to VADMs and calculate the mean and standard deviation as a function of N. A typical run is shown in Figure 1. From this simple simulation, it is evident that at least 25 paleofield estimates are required and that 18 might result in an biased result. Certainly, none of the individual studies has temporal sampling sufficient to provide a robust estimate of the paleofield during the CNS.
 The dearth of paleointensity data for the CNS stems from the difficulty of finding material suitable for the most reliable method of paleointensity determination, the Thellier-Thellier experiment [Thellier and Thellier, 1959]. The Thellier-Thellier experiment requires material whose magnetic phase is small enough to lack domain walls, yet large enough to retain a stable magnetization. Differences in cooling rate during acquisition of thermal remanence between the original and laboratory cooling strongly effects the ancient field estimate, so the ancient cooling rate must be reasonably well known, and if different from the laboratory cooling rate, the estimated field must be corrected [e.g., Selkin et al., 2000]. Furthermore, as the experiment entails repeated heating to progressively higher temperatures, the magnetic phase may not change its capacity to acquire a thermal remanence during the experiment. Submarine basaltic glass (SBG), formed when basalt is extruded into seawater and chills rapidly often is such a material [Pick and Tauxe, 1993a, 1993b, 1994].
 SBG is by nature highly susceptible to weathering. It can alter to hydrous phases and disappear from the geological record. Yet, ancient SBG can be found both in drill cores (e.g., from the Ocean Drilling Program [e.g., Selkin and Tauxe, 2000] and in outcrop [e.g., Robinson et al., 1983]). In this paper we exploit the fact that the Troodos Ophiolite formed during the CNS at about 92 Ma [Mukasa and Ludden, 1987] and has been exhumed, exposing remarkable outcrops of the ancient seafloor extrusives, including abundant submarine basaltic glass [Robinson et al., 1983; Rautenschlein et al., 1985].
2. Origin of Remanence in SBG
 The use of SBG for paleointensity measurements relies on a high temperature origin of its remanence. Heller et al.  challenged this, arguing that low Ti magnetite is not an equilibrium phase in MORB (citing Buddington and Lindsley ) and therefore it must be of low temperature origin. They propose that the primary high temperature Fe phase is Ti magnetite and speculate that the remanence in SBG is carried by low Ti magnetite that exsolves from high Ti magnetite after the glass has cooled down to 2°C. While clearly expressing that no such a process is presently known, Heller et al.  further speculate that SBG “probably contains a grain-growth chemical remanent magnetization in magnetite”. While Heller et al.  fail to offer supporting evidence for their proposed low temperature origin of SBG magnetization, there is a growing body of evidence that supports a high temperature origin.
 We know that the magnetite in the SBG forms very quickly as demonstrated by its presence in fresh material retrieved within a year of eruption. Glass samples from three separate flows whose eruption ages are very well known have been analyzed for paleointensity studies. Pick and Tauxe [1993a, 1993b] analyzed samples from the so-called “BBQ” site (9N on the East Pacific rise) erupted in 1990 which was sampled about a year later. Carlut and Kent  analyzed samples from two flows from the Juan de Fuca ridge whose eruptions were detected using sonic and seismic methods. The first (the “New Flow”) erupted in 1993 and was sampled within a month [Embley et al., 1995]. The second erupted in January 1998 and was sampled in July–September 1998 [Embley et al., 1999]. Fine grained basalts from the New Flow were analyzed by Kent and Gee . Carlut and Kent  also analyzed samples from the so-called “Animal Farm” Site on the Southeast Pacific Rise, but the eruption age is imprecisely known. All of these samples had magnetic properties typical of SBG, with blocking temperatures between 430 and 574°C, indicating low-titanium magnetite as the carrier of remanence. More importantly, the flows that had highly reproducible results (a requirement for a successful paleointensity experiment) yielded “paleofield” estimates in excellent agreement with the known field at the time of eruption.
 A second line of argument is based on kinetics. Magnetite begins demonstrably to grow when the samples are heated to above ∼600°C (Curie temperature experiments of Pick and Tauxe ), but does not seem to do so at ambient temperatures. The magnetizations since the samples were first acquired do not change significantly during storage - in other words, magnetite is not currently growing in the samples. Furthermore, grain growth CRM theoretically is about 0.4 of a TRM acquired in the same field [Stacey and Banerjee, 1974]. Therefore it seems hardly credible that magnetite would exsolve at 2°C within one month of eruption and acquire a grain-growth CRM that yields present field values from a paleointensity experiment.
 There is abundant petrologic evidence to suggest that low titanium magnetite forms as a high temperature phase in submarine basaltic glass [e.g., Zhou et al., 1997, 1999a, 1999b; Xu et al., 1997a, 1997b]. For example, [Zhou et al., 1997] studied basalts from the New Flow and found that while the large opaque oxides had the expected titanomagnetite composition (TM60), the interstitial glass had submicron titanomagntite grains with a broad range of titanium substitution (0 < x < 0.8). The discrepancy between the titanium substitution expected from equilibrium crystallization (cited by, for example, Heller et al. ) and that actually observed in interstital and rim glasses of freshly made glass can be reconciled by recognizing the fact that glass is not an equilibrium phase and arguments based on phase equilibria do not apply.
 The evidence therefore suggests overwhelmingly that low Ti magnetite in SBG can form as a high temperature phase, that it can evade alteration for geologically significant periods of time and that it can retain a primary thermal remanence yielding excellent paleointensity results. Juarez et al.  published preliminary data from 5 specimens from SBG obtained from the Troodos Ophiolite, suggesting that the material behaves well during the Thellier-Thellier experiment. On the basis of these arguments, we undertook an extensive sampling and laboratory program to exploit the CNS aged SBG exposed in the Troodos Ophiolite.
3. Sampling and Laboratory Analysis
 In April 2002, we sampled submarine basaltic glass (SBG) from 76 sites in four regions of the Troodos Ophiolite (see Figure 2). Glass was taken from the margins of sheet flows and intact pillows, dikes, and sills and from pillow margin breccias and glassy flows (hyaloclastites) (see Figure 3).
 Samples were first physically disaggregated with a mortor and pestle. This procedure is necessary to remove altered zones, but precludes the preservation of orientation. Individual pieces of glass up to 1 cm in width were washed in dilute HCl and examined under a binoculor microscope. Specimens that were free from signs of alteration, were unscratchable by a non-magnetic steel blade, and had net magnetic moments in excess of 1 nAm2 were placed in clean glass vials and fixed into position with microfiber glass filter papers. Thermal remanences of empty glass vials after heating to 600°C and cooling in a laboratory field (25 μT) were less than 10 pAm2.
 A total of 356 Troodos glass specimens were subjected to Thellier-Thellier type experiments [Thellier and Thellier, 1959]. Of these, 293 were treated as follows. Glass specimens were heated to 100°C and cooled in zero field. After measuring the natural remanent magnetization (NRM), the specimens were re-heated to 100°C, cooled in a laboratory field of either 25 μT or 40 μT, directed along the specimen's “Z” axis and remeasured. The difference in the z component between the second step and the first is the partial thermal remanence (pTRM) gained by cooling from 100°C to room temperature. At every other temperature step, there is an additional step between the zero field and the in-field steps: a reheating to a lower temperature step and cooling in the laboratory field. The pTRM gained in this repeated step (the so-called pTRM check of, for example, Coe et al. ), is compared with the first in field step at the same temperature. A difference is an indication of a change in the capacity of the specimen to acquire pTRM. After the in-field step, the specimens are heated for a third time to the same temperature and cooled in zero-field. This so-called pTRM tail check step [e.g., Riisager and Riisager, 2001] is used to test if the pTRM gained when cooling in the lab field is entirely removed by reheating to the same temperature step, a necessary condition for reliable paleointensity determinations. This zero-field, in-field (or ZI) heating procedure was repeated at 50 degree intervals up to 200 degrees and then at 25 degree intervals to the ultimate unblocking temperature of the NRM.
 The remaining 63 specimens were subjected to a slightly different experimental procedure referred to here as the IZZI protocol. At every other temperature step, the order of the double heating procedure (zero-field, in-field, ZI) is reversed such that specimens are cooled in-field first, then in zero field (IZ).
 All specimens that were subjected to the IZZI experiment were removed from the glass tubes and measured on a Princeton Measurements Corp. Micromag alternating gradient force magnetometer for magnetic hysteresis. All measurements were made at room temperature. In addition, at least two uncooked “sister” specimens from the same sites were also measured.
4.1. Thellier-Thellier Experiments
 The quality of the data from the Thellier-Thellier experiments can be assessed in a number of ways. Every paleointensity study encounters different behavior and must be judged in context. There are many parameters in the literature that describe behavior during the Thellier experiment, some of which we find useful here.
 Consider the data depicted in Figure 4. The points selected for the calculation of the best-fit slope (∣b∣) are solid circles and those omitted are open circles. The ancient field esitmate Banc is the absolute value of this slope times the laboratory field. The “scatter parameter” β is the standard error of the slope σ over ∣b∣ [e.g., Coe et al., 1978]. Coe et al.  also advocated the use of the parameter f which is the fraction of the NRM component used in the slope estimation. This is found by extending the best fit line through the chosen component as shown by the heavy dashed line in Figure 4. The parameter f is the ratio of the NRM fraction (vertical dashed line) and the component of the NRM (circled intersection). We find that while f works well with single component magnetizations, it is misleading when there are multiple components of remanence as in, for example Figure 4. The value of f for this specimen is quite high, whereas the fraction of the total NRM is much less. We prefer to use a parameter fvds which is the fraction of the total NRM, estimated by the vector difference sum (VDS) of the entire zero field demagnetization data (see Figure 4). The VDS “straightens out” the various components of the NRM by summing up the vector differences at each demagnetization step.
 The difference between the original pTRM at a given temperature step (horizontal component of the circles in Figure 4) and the pTRM check (horizontal component of the triangles in Figure 4), δi (see Figure 4), can result from experimental noise or from alteration during the experiment. Selkin and Tauxe  normalized the maximum δi value within the region of interest by the length of the hypotenuse of the NRM/pTRM data used in the slope calculation (the solid line through the selected data points). Their DRAT parameter is therefore the maximum Difference RATio. In many cases it is useful to consider the trend of the pTRM checks as well as their maximum deviation. We find therefore the sum of the differences more useful than the maximum δi. Here, we normalize this difference sum by the pTRM acquired by cooling from the maximum temperature step used in the slope calculation to room temperature. This parameter, expressed as a percentage, is called the Difference RATion Sum or DRATS.
 The pTRM tail check step is a rather new addtion to the Thellier experimental protocol family. The absolute value of the difference between the original NRM measured at a given temperature step (vertical component of the circles in Figure 4 and the second zero field step (known as the pTRM tail check) results from some of the pTRM imparted in the laboratory at Ti having unblocking temperatures that are different from Ti. The absolute value of these differences NRM (Δi) are plotted as squares in Figure 4. The Maximum Difference, normalized by the VDS of the NRM and expressed as a percentage is the parameter MD.
 In addition to parameters derived from the NRM-pTRM data, Pick and Tauxe [1993a] used two additional parameters associated with the directional variation of the NRM (see insert of Figure 4). These are the Maximum Angle of Deviation (MAD) of Kirschvink  of the selected component and the angle that that component makes with the origin. The latter we call the Deviation ANGle (DANG). MAD is a measure of the scatter in NRM directions and DANG tests whether the component selected is actually trending toward the origin.
 In certain specimens, the IZZI protocol leads to rather interesting behavior, described in detail by Yu et al. . The data with pTRM checks (associated with triangles) are the zerofield-infield (ZI) steps (lighter circles) and the intervening steps are the infield-zerofield (IZ) steps (darker circles). Alternating the two results in a “zigzag” in some specimens (barely discernible in Figure 4).
 We illustrate the use of our Thellier parameters and the IZZI protocol in Figure 5. The data shown in Figure 5a are typical of our best data. While many specimens behave like tg041j, 187 (52%) are problematic in at least one fundamental way.
 In Figure 5b we plot data showing a dramatic drift in pTRM checks at increasing temperatures. This is accompanied by a downward trend in the NRM directions (toward the lab field direction), caused by alteration of the specimen during laboratory heating. One danger with data like these is that there is a strong tendency to chose data from the low temperature portion of the NRM. Because these samples are Cretaceous in age, there is a likelihood that they have been contaminated with a VRM. For this reason, we use only components with maximum blocking temperatures of at least 350°C.
 Two examples of zig-zagging during the IZZI experiment are shown in Figures 5c and 5d. The zigzagging is strongly dependent on the direction of the NRM with respect to the laboratory field and is least evident when the NRM is perpendicular to the laboratory field (see Figure 5d which shows a strong zigzagging for tempertaure steps up to about 300°C while the NRM is vertical down (see inset) and no zigzagging when the field is more horizontal. Because of this directional dependence, it is hard to quantify the zigzagging effect itself. We note however that in general, specimens that display zigzagging have MD values of 4% or greater; these are excluded from further consideration here.
 In order to choose appropriate cut-off values for the rest of the parameters described above, we take a new approach. We plot the cumulative distributions of the various quality parameters in Figure 6. The cumulative distribution functions (CDFs) for these parameters tend to exhibit an inflection point shown as the solid vertical line. We use these inflection points as the criterion for acceptance as a convenient and reasonable way to identify the best data in this study. On the basis of these plots, we chose the following cut-offs for use in our paleointensity studies: DRATS ≤ 25; β ≤ 0.15; fvds ≥ 0.25; DANG ≤ 15; MAD ≤ 15. Furthermore, we require that there be at least two pTRM checks with blocking temperatures below the maximum blocking temperature chosen. Specimens passing all selection criteria are of grade “A”. Failure of any single criterion results in a grade of “B”, failure of two, a grade of “C” and so on. Of the specimens measured for this study, 178 were grade “A”. The rest have been eliminated from further consideration.
 All measurement data and their interpretations are available in the MagIC data base at http://earthref.org. By making all our data available, investigators with different preferences for selection criteria can re-analyze our data.
4.2. Hysteresis Measurements
 We summarize the hysteresis experiments in Figure 7. Representative loops are shown in the insets. The ratio of saturation remanence to saturation magnetization (Mr/Ms) is plotted against the coercive field (Bc) in the larger plot. These data range from highly coercive with high remanence ratios (e.g., tg001x, tg051i) to low coercive fields and remanence ratios (e.g., tg002i).
 Many of the loops are wasp-waisted (e.g., tg040j, tg001x) suggesting a mixture of single domain and super-paramagnetic grains [e.g., Tauxe et al., 1996]. The Thellier-Thellier data for specimens with wasp-waisted loops (see tg040j in Figure 5a) are generally of very high quality.
 Some loops (e.g., tg002i) are reminiscent of those produced by ensembles of magnetic grains whose remanence state is “vortex” [Tauxe et al., 2002]. Specimens with these vortex type loops appear to exhibit the most pronounced zigzag effect (see Figure 5d).
Smirnov and Tarduno  suggested that hysteresis measurements could be used to detect alteration of specimens during the Thellier-Thellier experiment. They also suggested that old submarine basaltic glass may suffer from undetected alteration. Specimens that altered significantly during our Thellier-Thellier experiment are circled in Figure 7. Unheated sister specimens are shown as plus signs. There does not appear to be a simple way to identify altered specimens from unaltered specimens on the basis of the hysteresis loops. Moreover, there is tremendous variety of behavior from multiple specimens from the same site so relying on “sister” specimens is risky.
 We argue that because hysteresis loops can never be performed on the same specimens as those used for Thellier-Thellier experiments until the experiment has finished and that because there is a tremendous variety of behavior from specimen to specimen that the hysteresis loops cannot be used to judge a particular experimental result. Moreover, the pTRM check procedure is designed to test specifically for alteration and, particularly when used in conjunction with the IZZI protocol, there is little chance that significant changes in the capacity for a specimen to carry a pTRM will go undetected.
4.3. Average Paleo Field From the Troodos Glasses
 We calculate site means for all sites with at least two estimates of paleointensity with a grade of “A” (Table 1). We have 39 sites meeting with standard deviations of less than 5μT or 15%. The grand mean of these is 33.6 ± 18.3 (1 σ) μT. If we assume a paleolatitude of the Troodos Ophiolite of 20° [see Juarez et al., 1998], we calculate a Virtual Axial Dipole Moment (VADM) of 75 ± 41 ZAm2.
Table 1. Site Means Using Specimens With Grade “A” Data (See Text)a
is average paleointensity. N is the number of specimens used in the calculation. σ is the standard deviation of . VADM is the conversion of to virtual axial dipole moment using an assumed paleolatitude of 20°.
4.4. Temporal Sampling of the Troodos Data Set
 We have obtained data from 76 sites, 39 of which meet optimal selection criteria. These come from Akaki Canyon and the Kampia mine region. The Akaki sites sample the lithologic units B-L of Rautenschlein et al.  and span nearly the entire extrusive thickness, including 11 cycles of volcanic activity, separated by 100s to 1000s of years). The rest of the successful sites are from the Kampia Mine region some 10 km to the East (see Figure 2). These also span the entire crustal thickness (excluding the top ∼50 m which everywhere are too oxidized). Depending on the assumed spreading rate (2–5 cm/yr), the two regions are likely to be different in age by 0.2 to 0.5 million years. We therefore estimate that our successful sites represent a temporal sampling of at least a few hundred thousand years of geomagnetic field variability.
4.5. Comparison With Other Data Sets
 We plot the updated compilation of paleointensity data in Figure 8 as squares and our new data as triangles. In an earlier compilation, Selkin and Tauxe  could find no significant difference between the period of low reversal frequency including the CNS and the early Cenozoic and the period of more rapid reversal frequency in the later Cenozoic. They noted however that there were very few data from the CNS itself.
 Since the compilation of Selkin and Tauxe , there have been several new studies which we have included in Figure 8 for a total of 18 published paleofield estimates. As noted in the introduction and illustrated in Figure 1, this is perhaps too few for a robust estimate of the paleofield and its standard deviation. In this study, we have more than doubled the number of data points in the CNS for a total of 57. It may now be possible to make a meaningful statistical comparison between the CNS and the post-CNS data. We show cumulative distributions of these two data sets in Figure 9. The grand mean of the 57 CNS sites in the compilation is 81 ± 43 ZAm2 or approximately equal to the present dipole field. This is significantly higher than the mean of the 180 post-CNS sites of of 56 ± 30 ZAm2.
 The result that the average VADM for the CNS is approximately equivalent to the present field resolves an apparent paradox. Long sedimentary paleointensity records suggested that the average strength of the field is related to the length of the polarity interval [Tauxe and Hartl, 1997]. The absolute paleointensity database rarely has sufficient data within a single polarity interval (with the exception of the Brunhes) to make the comparison between polarity interval length and average paleofield and Selkin and Tauxe  could establish no such relationship. The sedimentary record of DSDP Site 522 [Tauxe and Hartl, 1997] is currently the only published data set with sufficient numbers of paleofield estimates and variation in polarity interval length with which to test the relationship between polarity interval length and average field strength. It is therefore interesting to compare the data from Site 522 and the compilation of data from the CNS presented here.
 The relative paleointensity behavior for the 11 million year period of the Oligocene was published by Tauxe and Hartl  [see, also, Constable et al., 1998]. We convert the relative paleointensity estimates to quasi-absolute estimates of the geomagnetic field in μT using the technique of Constable and Tauxe  whereby average transitional field estimates are assumed to be 7.5 μT. In Figure 10 we show the VGP latitudes versus the relative paleointensity. The mean of the transitional data points (as identified in the magnetostratigraphy) is shown as a square. Normalizing this value to 7.5 μT and assuming a paleolatitude of 32°S for DSDP Site 522 in the Oligocene results in estimated VADMs for the DSDP 522 magnetic stratigraphy. In Figure 11a we plot the average VADMs versus polarity interval length and in Figure 11b we plot the same versus the standard deviation of the data. Also shown are the mean and standard deviation of the CNS data compiled here (stars). The predicted mean VADM for the CNS (∼40 million years long) agrees remarkably well with that observed.
 The results shown in Figure 11 suggest that when sufficient samples are available and when the paleointensity data are placed reliably within a Chron of known length, there is a relationship between polarity interval and paleofield strength. This observation supports the idea that strong fields inhibit reversals [Cox, 1968]. Furthermore, as noted by Constable et al.  the standard deviation of paleofield estimates is proportional to the average field is also supported by the CNS data.
 The average field during the CNS (when the field did not reverse) was the same as the present field. The average field for the post-CNS period (excluding the most recent field state of the last 1 million years) is much less. This observation would suggest that we are not heading for a reversal, but rather that the present field is unusually strong and the decrease in dipole field strength represents a return to the average, weaker state.
 1. The published literature contains 18 data points from within the CNS that meet minimum reliability criteria. Monte Carlo simulations from a statistical field model suggest that a minimum of 25 data points are necessary to obtain a robust estimate for mean and standard deviation of the paleofield strength.
 2. In this paper, we exploited the abundance of fresh submarine basaltic glass exposed in canyons dissecting the Troodos Ophiolite. Samples from 76 sites distributed all around the ophiolite were subjected to the traditional Thellier-Thellier experiment [e.g., Coe et al., 1978] or the new so-called IZZI experimental protocol, whereby temperature steps cooling in a laboratory field first, then cooling in zero field (IZ) alternate with zero field first, then in-field (ZI) steps. Data from IZZI experiments are useful for detecting subtle pTRM tails that are not demagnetized by the same temperature in which they were acquired resulting in a distinctive “zigzagging” in the NRM-pTRM plots.
 3. We chose critical cut-off values for various selection parameters by using cumulative distributions, an arguably less biassed method of setting selection criteria than the arbitrary way it is usually done. Of the 356 specimens analyzed here, 169 met these criteria. A total of 39 sites had at least two specimens with a standard deviation of ≤5 μT or 15%. The sites span at least several hundred thousand years and the average paleofield estimate is 33.6 ± 18.3 μT. Assuming a paleolatitude of 20 yields a mean dipole of 75 ZAm2, or approximately equal to the present dipole field.
 4. The data presented here more than double the data meeting minimum criteria: obtained from Thellier-Thellier experiments (with pTRM checks), at least two specimens per cooling unit with standard deviations of either 5 μT or 15%. Combining the data presented here with the published data results in a grand mean for the Cretaceous Normal Superchron of 81 ± 43 ZAm2 (N = 57), which is significantly larger than the mean of the post CNS data (84–1 Ma) of 55 ± 30 ZAm2 (N = 180).
 5. We converted the relative paleointensity estimates for the Oligocene magnetostratigraphic record of Tauxe and Hartl  to VADMs using the method of Constable and Tauxe . These data were then averaged by polarity interval. The plots of mean VADM versus polarity interval length and standard deviation predict values for the CNS in reasonable agreement with those presented here.
 6. The CNS appears to have been a period with an average dipole moment approximately equal to the present field. This observation suggests that perhaps the geomagnetic field is not racing toward transition as proposed by Hulot et al . Rather it could well be simply returning to a more average field state.
 We would like to acknowledge the cheerful field assistance of Daniel and Philip Staudigel. We thank Peter Selkin, Jeff Gee, Cathy Constable and Robert Parker for stimulating conversations. We especially wish to thank Agnés Genevey who co-developed the IZZI protocol used here. We are grateful to Jason Steindorf who made all the measurements. The manuscript was improved through the reviews of Scott Bogue, Mike McElhinny and Dennis Kent whose help we gratefully acknowledge. This work was supported in part by NSF Grant No. 0003395 to LT.