The loess-paleosol succession in the southeastern margin of the Chinese Loess Plateau retains high-resolution archives of sedimentary and environmental change. In this study, we report a detailed paleomagnetic and rock magnetic investigation of loess-paleosol sequences in the Sanmenxia area spanning the last 1.1 Myr. The results demonstrate that the Matuyama/Brunhes Boundary occurs at the top of soil S8 and the upper and lower boundaries of the Jaramillo Normal Subchron are encountered at the top of soil S10 and loess L13, respectively, in agreement with the classic Luochuan section. Loess L9, also referred to as “the upper sand layer,” is the coarsest loess unit over the last 1.1 Myr and carries an “imperfect” normal polarity magnetization. Although it is fully separated by a short reversed polarity interval, this thick normal polarity zone is too extensive to represent excursions like Kamikatsura and/or Santa Rosa and is concluded to represent an artifact rather than a record of geomagnetic field directions. Precise multiparameter environmental magnetic determination reveals that the short reversed polarity interval within the middle of this thick normal polarity zone exactly corresponds to a thin pedogenic horizon. Detailed rock magnetic analyses show that there are no intrinsic differences in magnetomineralogical properties between this pedogenic interval and other parts of L9, although the composition of magnetically soft components (magnetite/maghemite) in the short pedogenic interval is relatively higher than that of other parts of L9. Thus it is proposed that extensive remagnetization of parts of L9 most likely reflects a lithology-dependent process, and the dramatic coarsening of particle size across L9 is the fundamental reason for pervasive remagnetization. Our magnetostratigraphic results strongly imply that we need to reconsider the remanence-recording fidelity of coarse loess (L9, and probably L15) and care should be taken to interpret excursions in future paleomagnetic investigations of the Chinese loess.
 The exceptional success of magnetostratigraphy to identify major magnetozones and shorter subchrons from the Chinese loess has led to the desire for more detailed chronologies [Heller and Liu, 1982; Heller and Evans, 1995]. In recent years, a wealth of high-resolution magnetic investigations have been carried out on the extensive Chinese Loess Plateau (CLP) to detect polarity transitions, geomagnetic excursions, and to study acquisition processes of the natural remanent magnetization (NRM) in loess [e.g., Zheng et al., 1992; Zhu et al., 1994; Pan et al., 2002; Guo et al., 2002; Spassov et al., 2003; Yang et al., 2004]. However, the majority of magnetostratigraphic studies have focused on the hinterland and northwestern margin of the CLP, and investigations from the southeastern margin of the CLP have rarely been reported.
 According to the behavior of the East Asian paleomonsoon circulation, the loess accumulation thickness at the southeastern extremity of the CLP should be much smaller than at other areas of the CLP [An, 2000]. Nevertheless, a preliminary pedostratigraphic investigation demonstrates that well-exposed profiles in this area appear complete and paleosols easy to identify and separate [Yue and Xue, 1996; Zhao et al., 2000]. In this study we present a detailed magnetic investigation of a loess section in Sanmenxia with the intention to identify exact stratigraphic positions of main paleomagnetic polarity boundaries, assess fidelity of main reversal records and give a better understanding of the remanence acquisition process of the loess-paleosol couplets in this area.
2. Site and Sampling
 The 145-m-thick Sanmenxia loess section (34°38′N, 111°09′E) is situated at the southern bank of Yellow River, 15 km to the southwest of Shaanxian County, Henan Province (Figure 1). The section consists of 33 loess horizons and 32 soil horizons [Zhao et al., 2000].
 Oriented hand-samples (6 × 6 × 6 cm3) were collected at 10 cm interval from the top of loess L1 to paleosol S9 after removing weathered material. The topmost 1.4 m was not collected because of visible reworking by agricultural activity. In order to get the complete Jaramillo Normal Subchron (JNS) and Matuyama/Brunhes Boundary (MBB), the samples from the bottom of S9 to L13 and from L8 to the upper part of S8 were collected continuously. In the laboratory, three parallel sets of samples (named set A, B and C) were cut and a total of 2700 cubic specimens were eventually retrieved.
3. Experimental Methods
 Low-field magnetic susceptibility (χ, normalized by mass) was measured for all set A and B samples using a Bartington MS2B. All set A samples were then subjected to progressive thermal demagnetization (20–50°C intervals) to 580°C or 680°C using an ASC-TD48 thermal demagnetizer with an internal residual field lower than 10 nT. NRM and demagnetization measurements were made on a 2G-755 cryogenic magnetometer. The directions of the NRM components were determined with principal-component analysis using at least four temperature steps for each component. The anisotropy of magnetic susceptibility (AMS) was determined on selected set B samples using a KLY-2 induction bridge. Stepwise acquisition of isothermal remanent magnetization (IRM) was investigated using a solenoid to 180 mT succeeded by a Redcliffe pulse magnetizer to a maximum 3 Tesla. Thermomagnetic curves were run in an automatically recording, horizontal Curie balance by heating dried sediment in argon to 700°C (20°C/min) in magnetic field of 500 mT. Hysteresis measurements were performed on an alternating gradient magnetometer (Micromag 2900, Princeton). Anhysteretic remanent magnetization (ARM) was imposed in a 2G alternating field demagnetizer (160 Hz) in a DC field 0.1 mT and a maximum AC field 100 mT.
 Samples from the 8.9-m-thick loess unit L9 show rather complex demagnetization behavior as shown by selected demagnetization results (Figure 4). The majority of L9 samples (73 of 89 samples) display normal polarities, and most of them carry a single normal polarity component parallel to the present geomagnetic field (Figures 4a and 4b). A few samples carry two deviating normal polarity components, a low-temperature component erased at ∼200°C and a shallower high-temperature component prevailing to 580°C (Figure 4c). Only four samples clearly define a reliable reversed polarity overprinted with a significant normal-polarity direction (Figure 4d). Four samples carry scattered directions after 250°C preventing the determination of a reliable linear component. In stereographic projections, however, these samples show consistent up-pointing southerly directional distributions, suggesting a reversed polarity (Figure 4e). For eight samples, the demagnetization trajectories by-pass the origin, instead of decaying toward the origin (Figure 4f) with reversed inclination and northerly declination. The present results across L9 (Figure 3) are significantly different from previous reports on most Chinese loess profiles and in conflict with the well-established geomagnetic polarity timescale [Singer and Brown, 2002].
5. Rock Magnetic Properties of L9: Mineralogy and Grain Size
5.1. Lithology and Magnetic Susceptibility
 Loess unit L9 is readily identified in the field by its exceptional thickness, unique coarse texture, and paler color compared with other loess units [Guo et al., 1998]. The boundaries between L9 and the overlying and underling paleosols are clearly recognized by rapid shifts in χ values (Figure 3a). Careful field investigation combined with χ and other rock magnetic parameters confirms that L9 consists of three subparts: a coarser upper and lower part with a finer middle unit (Figure 5). Noticeably, the maxima of χ and frequency-dependent susceptibility (χfd%) associated with the fine-grained middle interval are comparable in magnitude to that of S7 and S8 and much more prominent than that of any other previously reported loess sections (e.g., Luochuan, Weinan, Baoji and Lingtai). This is regarded as evidence that the middle unit represents a distinct pedogenic interval. According to the standard stratigraphic nomenclature for the Chinese loess proposed by Kukla and An , the three second-order units of L9 correspond to L9LL1, L9SS1 and L9LL2, respectively, in order of increasing age.
5.2. Remanent Coercivity Analysis
 Progressive acquisition of isothermal remanent magnetization (IRM) in magnetic fields to 3 T shows that soil samples reach ∼90% of SIRM below 300 mT, indicative of a predominance of low-coercivity magnetite/maghemite (Figure 6). Loess samples reach SIRM in ∼400 mT, suggesting relatively higher coercivity phases. The gradual increase of IRM after field 300 mT can be attributed to the presence of high-coercivity minerals (hematite and probably goethite) for both loess and paleosols. The stepwise removal of the SIRM shows that paleosols have smaller coercivity of remanence (Hcr) with a mean value of 24 ± 4 mT while loess samples display a rather broad distribution of Hcr between 66 mT and 32 mT. The samples from L9LL1 and L9LL2 have greater Hcr values compared with the samples from L9SS1 and other loess units.
5.3. Thermomagnetic Analysis
 All loess and paleosol samples including those from L9 define the distinctive Curie-point of magnetite (580°C) (Figure 7). In accord with previous studies, loess samples display pronounced concave-down behavior at ∼300°C, reflecting the breakdown of thermally unstable maghemite, while this maghemite transition becomes less visible on the heating curves of paleosol samples. Specifically, the reduction in high-field induced magnetization after heating to 700°C is always greater in loess than in paleosols (Figure 7). This can be attributed to a more widespread conversion of magnetite/maghemite to hematite.
5.4. Hysteresis Analysis
 Representative hysteresis curves are shown in Figure 8. For most of the L9 samples, hysteresis loops are not closed until around ∼500 mT, and display a weakly developed wasp-waisted shape. This wasp-waisted behavior is much more pronounced for the samples from L9SS1 and its neighboring soil units (Figure 8), probably indicating increasing contribution of pedogenic fine-grained fraction. In the Day plot, all samples fall in the pseudosingle-domain range with insignificant differences in the ratio of Mrs/Ms (Figure 9). The samples from L9LL1 and L9LL2 exhibit greater Hcr/Hc values compared to paleosol samples, probably suggesting somewhat coarsening of magnetic grains in the coarser units of L9.
5.5. Three-Component IRM Demagnetization
 Thermal demagnetization of the orthogonal, three-component IRM yields demagnetization curves controlled by both coercivity and thermal unblocking temperatures of potential magnetic carriers. We imposed orthogonal IRMs on 16 samples from L9 and its neighboring loess/paleosol units using the technique proposed by Lowrie . Fields of 2.2 T, 0.4 T and 0.15 T were successively applied in orthogonal directions. Figure 10 shows that the IRMs are mainly controlled by a low-coercivity component unblocked at about 580°C, reflecting the predominance of magnetite. The pronounced change in the slope of the soft-component curves between 250°C and 350°C probably indicates the breakdown of maghemite, in accord with results from thermomagnetic analysis. Both the hard and medium coercivity components reveal the presence of hematite with unblocking temperature of 680°C. The results also show a relatively higher proportion of hematite in L9 than in S8 and S9 (Figure 10).
6. Anisotropy of Magnetic Susceptibility
 In order to detect whether sampling or reworking after sedimentation have distorted the paleomagnetic record, the anisotropy of magnetic susceptibility (AMS) was determined for all L9 samples. Figure 11 shows the maximum (Kmax) and intermediate (Kint) axes of AMS are intermixed in the horizontal plane, while the minimum axes (Kmin) are subvertical to the bedding plane (Dec = 29.0°, Inc = 86.8°, α95 = 2.0). This magnetic fabric is characterized by foliation-dominated ellipsoids and indicative of a deposition-related magnetic fabric without structural deformation and/or obvious distortion by pedogenesis.
7.1. Kamikatsura/Santa Rosa Excursions?
 L9 consists of three subunits: the coarsest L9LL1, finer L9SS1 and coarser L9LL2 (Figure 5). The four samples with reversed polarity are all confined to the 40 cm thin L9SS1 subunit, dividing the normal polarity L9 by a thin reversed polarity interval coinciding with distinct changes in lithology. Moreover, the lower part of the normal polarity (corresponding to L9LL2) contains several samples displaying reversed polarities (Figure 4e). On the basis of the timescale given by Heslop et al. , L9 spans 87 kyr implying an average accumulation rate of L9 in the Sanmenxia of the order of 10.2 cm/kyr. Recent high-resolution dating of short geomagnetic excursions from igneous rocks and marine sediments supports the existence of two normal-polarity events, e.g., Kamikatsura and Santa Rosa, between the MBB and JNS [Singer and Brown, 2002]. However, both Kamikatsura and Santa Rosa around 890–900 ka and 930–940 ka, respectively, are very short events probably spanning less than 10 kyr. All samples within the 5.3-m-thick L9LL1 define normal polarities. Hence it is unrealistic to assign the whole normal polarity interval or the upper part of the normal-polarity interval to the Kamikatsura or Santa Rosa event. Considering that Kamikatsura and Santa Rosa are separated by a fully reversed polarity interval spanning some 37 ± 9 kyr and only four samples from L9SS1 clearly define a very short reversed polarity interval, we rule out the possibility that the upper and lower part of the normal polarity interval represent the Kamikatsura and Santa Rosa excursions.
 Detailed rock magnetic analyses demonstrate that the rock magnetic characteristics of L9 are consistent with numerous rock magnetic reports on other parts of the CLP [e.g., Zheng et al., 1992; Heller and Evans, 1995; Guo et al., 2002; Yang et al., 2004]. Thus we propose that the short reversed polarity interval has retained a “primary” record of the geomagnetic polarity more or less at the time of deposition/pedogenesis while the normal polarity intervals of L9 represent complete remagnetization.
7.2. Evidence From the Hinterland of the CLP
 Several magnetostratigraphic studies across the CLP suggest the existence of one or more short normal polarity intervals in the upper part of L9. Sections in Luochuan and Baoji reveal a few normal polarity data points between the MBB and the JNS [Liu and An, 1984; Rutter et al., 1991; Yang et al., 2004]. At Xifeng, a more complex polarity pattern is reported showing frequent polarity flips within L9 [Liu et al., 1987]. Short intervals of normal polarity characterized by a southerly swing to westerly declination with shallower reversed inclination to normal inclination have been reported from Lantian [Zheng et al., 1992] and Pingliang [Sun et al., 1998]. Recently, a study from Fanshan in the Beijing region also defines a normal polarity interval within L9, the thickness of which reaches 4∼5 m [Xiong et al., 2001]. Some conspicuous results from Lanzhou by Rolph et al.  compelled them to assign two normal-polarity intervals within L9 to the JNS and Cobb Mountain event, respectively. Nevertheless, none of the reported normal polarity “events” within L9 is as thick as that of Sanmenxia, and they have commonly been attributed to unsuccessful removal of magnetic overprints [Heller and Evans, 1995].
 Detailed magnetostratigraphic investigation across L9 by Yue and Xue  confirmed the existence of normal polarity directions within sections from Duanjiapo, Jiuzhoutai, Pingliang, Xifeng and Luochuan, and they proposed it reflected the presence of a “Lantian” excursion. Noticeably, the first results from the classic Luochuan section published by Heller and Liu  are also not readily interpretable in terms of the standard geomagnetic polarity timescale because some data points apparently possess “wrong” polarity including those from the upper part of L9. They suspected that some samples had been misoriented during sampling (turned upside-down), and then performed corrections based on directions of secondary overprints widespread in the CLP [Evans and Heller, 2001]. Also, the MBB was wrongly located within “the lower sand layer”-L15 in the earliest magnetostratigraphic dating of the Chinese loess deposits [An et al., 1977]. This can be attributed to insufficient sensitivity of the measuring equipment, lower sampling resolution and inadequate demagnetization treatment. However, extensive remagnetization of parts of L9 (and probably also L15) is probably the fundamental reason that some samples do not respond to laboratory cleaning.
7.3. Origin of Remagnetization
 The pervasive remagnetization of L9 may be caused by several processes:
 1. Viscous remanent magnetization (VRM). Similar to previously numerous paleomagnetic results from the hinterland of the CLP [Zheng et al., 1992; Zhu et al., 1994; Guo et al., 2002; Yang et al., 2004], most Sanmenxia samples (except those of L9) display a well-defined ChRM after removal of a low-temperature overprint by peak temperatures between 200°C and 250°C (Figure 2). This low-temperature component of magnetization directed along the present geomagnetic field could be attributed varying degrees of low-temperature oxidation of stoichiometric magnetite during the first stages of loess deposition [van Velzen and Dekkers, 1999] or a VRM [Pan et al., 2001]. For most L9 samples, a single normal polarity component parallel to the present geomagnetic field persists up to 580°C or 680°C (Figures 4a and 4b), despite deposition during a time of presumed reverse polarity. According to the Néel's single domain theory, a VRM acquired over the duration of Brunhes at 20°C should unblock during thermal demagnetization at about 175°C [e.g., Pulliah et al., 1975]. Thus it is unlikely that such a stable normal-polarity component is a VRM. Theoretical considerations suggest that coarse multidomain magnetic grains are not good preserver of a stable remanence since they contain rather easily moved domain walls subject to the internal self-demagnetizing fields [Dunlop and Özdemir, 1997]. However, Day plot suggests that the systematic coarsening of the loess material of L9 does not result in significant coarsening of magnetic mineral grains in L9 (Figure 9).
 To give a better estimation of VRM, eleven fresh samples from L8, S8, L9 and S9 placed in a low-field cage (Hamb < 10 nT) for 30 days were subjected to a NRM decay experiment. For all samples, the linear relationship between NRM remaining and elapsed time on a logarithmic scale generally reflects a relaxational VRM (Figure 12). This short-term VRM-decay experiment also indicates that during this interval some 20% of the NRM was lost for samples from L9LL1 and L9LL2 and 25 to 35% lost for those from L9SS1. The paleosol samples from S8 and S9 also exhibit much greater viscous loss compared to those from L8 and L9, which is consistent with previous interpretation that VRM is dominantly carried by fine-grained magnetic particles approaching the single domain and superparamagnetic boundary [Pan et al., 2001]. Since the VRM decay of the NRM is largest in the paleosol samples, this would suggest, these are likely to be most dominated by VRM acquisition, hence a VRM contribution to the loess remagnetization is likely to be less than the paleosol samples. Thus it appears that VRM is not the actual process of the inferred L9 remagnetization.
 2. Acquisition of a chemical remanent magnetization (CRM) promoted by pedogenesis and weathering. Pedological evidence demonstrates that the L9 unit, ascribed as “the upper sand layer,” is the coarsest loess layer and is inferred to represent the severest glacial conditions during the last 1.1 Myr. This loess has consequently experienced the weakest weathering and pedogenesis [Guo et al., 1998]. Remanent coercivity, hysteresis and thermomagnetic analysis reveal that there are no significant magnetomineralogical differences between L9 and other loess units. The relatively greater Hcr values, less pronounced wasp-waisted behavior and much lower χfd% values of L9 compared with other loess/paleosol units somewhat exhibit rock magnetic properties of unaltered loess. An average χ of L9 (excluding L9SS1samples) of 36.51 × 10−8 m3 kg−1 is in accord with values of detrital loess, indicating no major chemical precipitation/alteration of magnetic minerals in the loess of L9. In contrast, the thin L9SS1 horizon as reflected by mineral magnetic parameters (Figures 3a and 5) marks a short “spell” of warmer climate. The four reversed polarity samples from L9SS1, all samples from S9 and L10, and all S8 samples below the MBB must have experienced much stronger pedogenesis/weathering than L9LL1 and L9LL2. Figure 3 clearly shows all these samples unambiguously carry reversed polarity acquired during the Matuyama chron while the least pedogenized L9LL1 and L9LL2 apparently yield “wrong” polarity. It is therefore tentatively proposed that a CRM induced by pedogenesis cannot account for the complete overprinting of the original magnetization of L9.
 3. Postdepositional remanent magnetization (pDRM) acquired by physical realignment of magnetic grains. L9 consists of 45–60% coarse silt (25∼50 μm) [Guo et al., 1998] and is consequently susceptible to postdepositional realignment of magnetic particles in the ambient geomagnetic field upon drying or wetting [Verosub, 1977; Payne and Verosub, 1982; Løvlie, 1994]. According to the behavior of the East Asian monsoon circulation [An, 2000], the Sanmenxia area must have experienced the strongest summer monsoon intensity and have a higher mean annual precipitation compared to other parts of the CLP [Guo et al., 1998]. Thus during interglacials succeeding S8, the coarse-grained nature of L9 may have acted as a conduit for water percolation.
 Demagnetization results and mineral magnetic parameters suggest that the proposed remagnetization of L9 depends on grain size (Figures 3 and 5). This is evident from the observation that all samples from the coarsest part of L9LL1 and most samples from the coarser interval of L9LL2 have been remagnetized, while samples from the finest L9SS1 clearly define reliable reversed polarities. Therefore the inferred extensive remagnetization of parts of L9 probably reflects preferred postdepositional realignment by wetting during the Brunhes, and then hence samples retain a pDRM by physical reorientation of magnetic mineral grains parallel to the ambient geomagnetic field. For a few L9LL2 samples which do not carry a consistent stable remanence with unblocking temperature above 200°C (Figures 4e and 4f), three-component IRM demagnetization results suggest that the reason could not be plausibly attributed to the lack of thermally stable remanence carriers and/or predominance of VRM, but probably reflect randomization of the magnetic particles during reorientations and hence inhomogeneities of paleomagnetic directions because of limitation of rotatable space for smaller ferrimagnetic grains. Moreover, several samples define two deviating components with normal polarities, a viscous overprint erased at 200°C and a high-temperature component that may represent partial remagnetization of all magnetic grains. However, the physical reality of this conceivable scenario needs to be experimentally verified by proper redeposition investigations to further improve and ultimately establish the remagnetization mechanisms of coarse loess units.
 A detailed magnetostratigraphic investigation at the southeastern extremity of the CLP reveals a thick normal polarity interval between MBB and JNS that is too thick to be assigned as an excursion like Kamikatsura/Santa Rosa but represents an artifact rather than true geomagnetic directions. Detailed rock magnetic analyses indicate that this normal polarity interval is lithology-dependent, and the systematic coarsening of particle size across L9 is probably the fundamental reason for pervasive remagnetization. We tentatively propose that remagnetization of parts of L9 was accomplished by physical rotation/reorientation of the remanence-carrying grains at burial depths of 9 meters or more, some 100 kyr after initial deposition.
 Our magnetostratigraphic results combined with previous paleomagnetic investigations from the hinterland of the CLP suggest that remagnetization of parts of the Chinese loess may be more widespread than previously anticipated. Consequently, interpretation of short normal polarity intervals in coarse-grained loess units should be treated with caution in future paleomagnetic investigations.
 We are indebted to F. C. Jiang for his encouragement and suggestions throughout this study. Thorough and thoughtful reviews by M. Jackson, M. Hounslow, and the Associate Editor C. Constable substantially improved the clarity and scope of this manuscript and are greatly appreciated. We thank C. Kissel at LSCE/CEA/CNRS, Gif-sur-Yvette, for the opportunity to run hysteresis curves. This work is supported by the National Natural Science Foundation of China (grant 40202018), China Geological Survey (grant 200413000035), and the Ministry of Personnel of China.