We report the results of several surface geophysical surveys of the central part of the Mariana Basin, western Pacific. Analysis of bathymetric and magnetic data along three distinct spreading segments indicates highly asymmetric spreading processes, the half-spreading rate being two to three times more important on the western side of the basin than on its eastern side. Such an asymmetry is as strong at the centers of segments as at their ends, and does not systematically correlate with an asymmetric shape of the rift valley. A half-graben shape of the rift valley is observed only at segment ends. At segments centers, a symmetric graben shape characterizes the valley. The asymmetric shape of the valley at segment ends probably reflects the localization of the tectonic strain due to changes in lithospheric rheology and thickness. Asymmetric spreading is likely accomplished through a different amount of deformation and magma emplacement on both sides of the spreading axis. It may result from the combined effects of (1) asymmetric rheologic, melting and stress conditions imposed by the influence of the pre-existing volcanic arc and its melt sources on the eastern side of the basin, (2) the northwestward movement of the Philippine Sea Plate while the almost vertical Pacific slab in the Mariana subduction zone resists to this motion, (3) a roll-back component at the Mariana subduction zone.
 The Mariana Basin is currently opening for 6 Ma [Hussong and Uyeda, 1982] behind the Mariana subduction zone, where the Pacific Plate subducts beneath the Philippine Sea Plate [Seno et al., 1993] (Figure 1). It extends from 12°N to 23°N and is characterized by a crescent shape. Thickness of the crust in its central part varies from 4 to 7 km, and its velocity structure is typical of slow-spreading ridges crust [e.g., LaTraille and Hussong, 1980; Sinton and Hussong, 1983]. Where not covered by sediments, the seafloor is deeper and the bathymetry more rugged than a normal slow spreading crust [Park et al., 1990; Stern et al., 2003]. Magmas in the central part of the basin are back-arc basin basalts (BABB). They are very similar to normal mid-ocean ridge basalts, yet they are affected by a “subduction component” that travels with fluids extracted from the subducted slab or sediments [Newman et al., 2000; Stern et al., 2003].
 In the central part of the Mariana Basin, the morphology of the spreading axis is almost similar to the one of slow-spreading mid-ocean ridges [Stüben et al., 1998]. It is composed by several spreading segments, each being 20 to 50 km in length. Their morphology varies from wide, deep valleys with or without a median axial volcanic ridge, to narrower, shallower and hourglass-shaped ones (similar to the rifted axial highs of intermediate-spreading ridges), all at slow spreading rate [Seama and Fujiwara, 1993; Seama et al., 2002]. Segments with such hourglass-shaped morphologies are interpreted as being more magmatically robust [Fox et al., 1991; Sempere et al., 1993]. Within a single spreading segment, the axial depth varies by 1–2 km between the shallow axial high at the center of the segment and the deeper segment ends. Moreover, each segment whose length exceeds 20 km displays along-axis variations in crustal thickness and/or upper mantle density, as demonstrated by low residual gravity anomalies at segment midpoints [Kitada et al., 2002]. This is seen as an evidence that magma emplacement is focused at the depth minimum of individual segments [e.g., Macdonald, 1998; Sempere et al., 1993].
 Several surface geophysical surveys have been conducted in the Mariana Basin since 1992. In this paper, we present bathymetric and magnetic data over three segments (A, B, and C) of the central Mariana basin's spreading axis. Segments A and B are located between ∼17°40′N and ∼18°28′N (Figure 2a). Magnetic data were collected during the KH92-1 cruise onboard the R/V Hakuho-Maru [Kong, 1993; Seama and Fujiwara, 1993]. Bathymetric data were collected during during the YK99-11, YK00-13, and YK01-11 cruises onboard the R/V Yokosuka. Segment C (Figure 2c) is located between ∼19°20′N and ∼19°50′N. It was surveyed during the KR98-12 cruise onboard the R/V Kairei [Yamazaki et al., 1999].
 Bathymetric data constrain the morphology of the rift valley and the characteristic of faults on its both sides. Magnetic anomaly was calculated by subtracting the International Geomagnetic Reference Field (IGRF) 1990 and 1995 models [International Association of Geomagnetism and Aeronomy (IAGA) Division V Working Group 8, 1991, 1995], for 1992 and 1998 data sets, respectively. Crustal magnetization was calculated from measured magnetic anomaly (location of data on Figure 3). Primary purpose of the magnetic inversion is to remove the skewness due to low magnetic latitude to obtain a better estimate of the source magnetization locations, their relative magnitudes, and their relation to features on the seafloor. The inversion method developed by Parker and Huestis , and later extended to three-dimensions by Macdonald et al. , was used for this calculation. The data were gridded in 0.5 arc min cells. A uniform, 500 m thick magnetic source layer was assumed. The direction of magnetization in the source layer was assumed to be oriented parallel to a geocentric dipole field. Bandpass filters with cosine tapers for wavelengths between 3–6 km and between 100–150 km prevented instabilities during the inversion. No annihilator was added for the magnetization calculation in segments A and B area (location on Figure 2a). Because the data coverage is limited in this area, there is inherent ambiguity in the zero level of the magnetization. However, an inflection point of maximum slope is placed at the zero crossing of the across-axis profile of magnetization (Figure 4). The inflection point is supposed to point the close vicinity of the edge of Brunhes chron according to a forward modeling [e.g., Vacquier, 1972]. The annihilator was added once for the calculation in segment C area to give approximately equal positive and negative magnetization values (Figure 4).
3. Morphologic and Magnetic Observations
 The segment A is ∼55 km long, trends N15°W, and is bounded by non-transform discontinuities at ∼17°55′N and ∼18°28′N (Figures 2a and 2b). The median valley floor consists of a clear and wide (10–15 km) graben bordered by rectilinear walls in which large normal faults (throws range from 100 m at the segment center to 1400 m at its end) displace the crust upward to form the crestal mountains. The valley floor shoals to <3700 m in depth toward the segment center (at ∼18°13′N), and deepens to >4700 m and slightly widens toward its end. It contains an almost linear and apparently unfaulted axial volcanic ridge (AVR) (1–5 km wide, up to several hundreds of meters high) that likely marks the locus of the most active recent accretion within the neovolcanic zone [e.g., Ballard and van Andel, 1977; Smith et al., 1999]. At the vicinity of the segment center, east of the main AVR, shorter (2 to 3 km -long) hummocky ridges may represent other sites of lava extrusion (Figures 2a and 2b). At segments ends, the shape of the rift valley is asymmetric with higher, steeper walls on its eastern side, and a gentler slope to the west (Figure 4b). At segment center, the valley is roughly symmetric (Figure 4a).
 Magnetic data indicate that spreading occurs at a full rate of 31 to 35 mm/yr along segment A. The central magnetization peak is centered on the AVR (Figures 4a and 4b). The boundaries of the positive magnetization high (interpreted as the Brunhes anomaly that characterizes seafloor younger than 0.78 Ma) are not centered on the AVR. At the segment center, the edge of the central Brunhes anomaly is 13 km and 11 km from the AVR and thus the half-spreading rate is 16.6 and 14.1 mm/yr on the western and the eastern flank, respectively (Figure 4a). If the short hummocky ridges located east of the main AVR (Figures 2a and 2b) rather represent the most recent locus of volcanism, the spreading rate would be 19.8 and 11.0 mm/yr in the west and in the east, respectively. At the southern end of the segment, the asymmetry of the Brunhes anomaly with respect to the locus of volcanism is pronounced (Figure 4b). Its edge is 15 km and 10.5 km from the AVR on the western and the eastern flank, respectively. The half-spreading rate is then 19.2 and 13.8 mm/yr at the western and the eastern half-plate.
 The segment B is 25 km long (Figures 2a and 2b). It is bounded by non-transform discontinuities at ∼17°40′N and ∼17°55′N. It overlaps segment A to the north (Figure 2b). The morphologic expression of the AVR is not clear. Depending on the location along the segment, we notice the presence of one to two apparently unfaulted ridges that are composed of seamounts and terraces, and thus may both represent the locus of current volcanism. With regard to its morphology, the axial valley is asymmetric due to the presence of an elevated inside corner high (IC) on its western side (Figures 2a, 2b, and 4c). The full spreading rate is 30.8 mm/yr at its center. The edge of the central Brunhes anomaly is 15 km and 9 km on the western and the eastern flank, respectively, from the most developed and continuous AVR that coincides with the peak of the positive magnetization. The half-spreading rate is thus 19.2 and 11.5 mm/yr at the western and the eastern half-plate, respectively.
 The segment C is delimited by non-transform discontinuities at ∼19°20′N and ∼19°50′N (Figures 2c and 2d). It is characterized by a general dome-shaped morphology. Along this segment, the median valley floor has an hourglass shape, narrowing and shoaling to <3700 m in depth toward the magmatic center (at ∼19°37′N), and widening and deepening to >4800 m at its end. The valley generally does not contain an axial volcanic ridge, suggesting that the site of lava extrusion is distributed within the floor of the valley (Figures 2c and 2d). Spreading occurs at a full rate of 20 to 25 mm/yr. Near the segment center, the median valley is not well marked (Figure 4d). The frequency and throw of faults on its both sides are generally equal. Despite the symmetric shape of the valley, the central positive magnetization high is strongly asymmetric, with its highest peak centered on the valley axis, and its western and eastern boundaries being located 19 km and 3 km from this axis, respectively. The half-spreading rate is 17.9 and 6.4 mm/yr at the western and the eastern half-plate, respectively. At the southern end of the segment, the shape of the rift valley is highly asymmetric (Figure 4e). It consists of a half-graben with several small throw faults at its western side, and a single to double large fault scarp on its eastern side. The central positive magnetization high is also asymmetric. From its maximum at the center of the axial valley, magnetization decreases sharply to the east over only 5 km toward the large valley wall fault, but gradually westward over 11 km. The half-spreading rate is 14.9 and 5.1 mm/yr at the western and the eastern half-plate, respectively (Figure 4e).
4. Relationships Between Asymmetric Accretion and Tectonic Features
 Within the three studied spreading segments, the central magnetization high that characterizes the Bruhnes anomaly is systematically asymmetric with respect to the inferred locus of volcanism that coincides with the peak of the central positive anomaly and with either the axial volcanic ridge (AVR) or the valley axis. The steepest magnetic gradients are always observed on the eastern side of the spreading segments, and often next to one or few large inward-facing normal faults at segment ends. This is the case of the southern ends of segments A and C (Figures 4b and 4e). The northern ends of these segments are also characterized by such an asymmetric half graben but the lack of accurate magnetic data prevents any reliable study of the spreading processes at these locations. We cannot observe along-axis morphologic variations of segment B in cross -section due to its limited length. Segments B and C display a highly asymmetric Brunhes anomaly both at centers and ends. Segment A displays a highly asymmetric Brunhes anomaly at its southern end, and slightly to highly asymmetric magnetic anomaly at its center, depending on the locus of volcanism considered. In general, the asymmetric magnetic pattern observed along the central Mariana spreading segments reveals asymmetric spreading processes that occur within either an asymmetric or symmetric rift valley.
 Asymmetric spreading has been reported from the slow-spreading Mid-Atlantic Ridge within the “FAMOUS” segment (36°N) [e.g., Macdonald, 1977], the “Broken Spur” segment (29°N) [Searle et al., 1998; Allerton et al., 2000], and at 45°N [Loncarevic and Parker, 1971], and 26°N [McGregor and Rona, 1975]. Except for the case of the FAMOUS segment [Macdonald, 1977], the asymmetry of the spreading is generally slight at segment centers, and more developed at segment ends [Allerton et al., 2000]. The segment A in the Mariana Basin may display such characteristics if we consider that its longest hummocky ridge represents the current locus of volcanism. Spreading along segments B and C is highly asymmetric both at the center and the end of segments. At the center of segment B, the spreading rates on opposite sides of the spreading axis differ by a factor of two along the whole segment. The segment C has the highest degree of asymmetric spreading ever reported from the center of a slow spreading segment, with nearly a factor of three difference between the half-spreading rates on opposite sides.
5. Mechanism of Asymmetric Spreading
 It is has been observed that asymmetric spreading occurs at locations where the axial valley is highly asymmetric in cross section: the slow-spreading side of the valley displays an inner wall that is a single major step composed of a series of steeply faulted slivers, whereas the fast one has an inner wall consisting of several wide steps, less steeply faulted and with a more gradual average slope [Macdonald, 1977; Searle et al., 1998; Allerton et al., 2000]. The density of faulting is nearly twice as great on the fast side of the valley but the fault throw is generally limited [Macdonald, 1977]. According to Searle et al.  and Allerton et al. , spreading on the fast side is thought to occur through magmatic accretion of upper crustal material (layer 2 composed of basalts and gabbros), while spreading on the slowest side is accommodated mainly by significant displacement on a single, large fault. Such an asymmetric accretion is accomplished by progressively shifting the locus of volcanism to a new location at the slowest side of the valley. The locus of accretion is likely to be a region of local extension mainly controlled by flexure associated with the large fault that bounds the slowest side of the valley [Allerton et al., 2000].
 The above mechanism of asymmetric spreading may be valid concerning the portions of the Mariana spreading segments that display a clear half-graben shape of the rift valley. This morphology might indeed indicate that within the Mariana spreading axis, new crust is preferentially added to the western plate, while extension is rather accommodated by tectonic faulting, i.e., significant displacement along one or few large faults scarps on its eastern side. However, in our case, highly asymmetric accretion is also observed at the centers of segments B, C and possibly A, where the axial valley displays a symmetric shape and an apparent symmetric distribution of faults (throw, spacing) on its both sides. At these locations, the model mentioned above is no more valid because there is no large fault on the eastern flank of the valley, probably because such large faults may be not sustained by a weak lithosphere toward midpoints of ridge segments [Escartin and Lin, 1995].
 From a study of detailed deep-tow sonar data over a spreading segment of the Mid-Atlantic ridge at 29°, where asymmetric spreading is evidenced by Allerton et al. , Escartin et al.  observe a marked asymmetry in tectonic strain of up to 50% on both flanks of the segment (at its center as well as at its ends), likely related to asymmetric magmatic accretion. It is worthwhile to mention that they observe that along-axis differences in fault geometry (heave, spacing) and in valley morphology (symmetric to asymmetric graben at segments center and ends, respectively) do not directly correlate with variations of the tectonic strain, but instead likely reflect along-axis changes in lithospheric rheology (thickness, composition): at segment ends where the magma supply is reduced and the lithospheric thickness higher, the amount of tectonic strain is the same as in the segment center, but it localizes to form large rift-bounding faults that accommodate most of the strain early in the history of the oceanic crust [Escartin et al., 1999].
 Similarly to the Mid-Atlantic ridge at 29°N, the asymmetry of the spreading processes at the Mariana segment centers might be accommodated by different amounts of tectonic strain and magmatic accretion on each side of the spreading axis. The absence of detailed deep-tow sonar and bathymetric data prevents us to perform a reliable calculation of the tectonic strain on both sides of the axis. However, residual gravity anomaly lows that systematically locate slightly to the west of the spreading segments in the study area [Kitada et al., 2002] may indicate that magmatic processes are more important on the western side of the axis than on its eastern side. The occurrence of large normal faults at segment ends on the eastern side of the Mariana rift valley does not demonstrate that the tectonic strain is larger at these locations, according to the recent studies of the Mid-Atlantic ridge [Escartin et al., 1999]. This indeed may simply reflect the presence of a weak (due to the presence of serpentinites) and thick (due to low temperatures) lithosphere at segment ends (segments A and C ends) or at magmatically non-robust segment centers (segment A center), and then the localization of the tectonic strain along a large bounding fault, resulting in an asymmetric shape of the rift valley at these locations. At the center of magmatically robust segments (segments B and C), where the brittle layer is likely thin and serpentinites scarce, the tectonic strain is rather accommodated by normal movement on closely spaced faults with small throws. It is worth noting that in this case, asymmetric accretion most likely cannot be sustained through a displacement of the locus of accretion related to the local extension due to bending of one large fault bounding the spreading axis as proposed by Allerton et al. .
 Detailed bathymetric and magnetic data collected close to the seafloor along asymmetrically spreading segment are needed to resolve the question of the mechanisms of asymmetric spreading in the Mariana Trough. They will indeed be useful to investigate the distribution of faults and volcanic edifices within the axial valleys, and to measure precisely the morphological characteristics of small-size faults such that reliable calculations of tectonic strain will be possible.
6. Origin of Asymmetric Spreading
 The pattern of spreading in the Mariana Basin may be related to (1) local features imposed by back-arc setting, (2) the northwestward movement of the Philippine Sea Plate while the almost vertical Pacific slab in the Mariana subduction zone resists to this motion, and/or (3) roll-back component at the Mariana subduction zone:
 1. It has been shown that back-arc basins tend to evolve highly asymmetrically from their rifting stages to more symmetric spreading [e.g., Barker and Hill, 1980; Martinez et al., 1995; Parson and Wright, 1996; Yamazaki and Murakami, 1998; Martinez and Taylor, 2002; Schellart et al., 2002], due, at least in part, to asymmetric rheologic, melting and stress condition imposed by the influence of the pre-existing volcanic arc and its melt sources, which are linked to the mantle wedge and slab [e.g., Martinez and Taylor, 2002]. These conditions could result in the frequent relocation of the spreading center close to the active volcanic arc. In the case of the central Mariana Trough, they might influence the pattern of spreading only slightly since (1) the spreading axis is morphologically similar to mature spreading centers, (2) it is located at about 100 km west from the active arc, and (3) it is characterized by lavas that are back-arc basin basalts (BABB) [Newman et al., 2000; Stern et al., 2003]. Integrated seismic and electro-magnetic experiments will bring new constraints that are necessary to validate this hypothesis.
 3. The lack of geometric fit between the remnant West Mariana remnant arc and the present-day active Mariana Arc which has a greater curvature [Karig et al., 1978] (Figure 1), and the pattern of active or recent faulting in the Mariana forearc at high angles to the trench [Stern and Smoot, 1998; Martinez et al., 2000] may be related to a radial expansion of the arc as proposed by Karig et al. . This cannot be explained by rigid-plate rotation of the PSP away from a fixed-shape trench and then reflects the existence of a non-negligible roll-back component at the Mariana Trench.
 Finally, the pattern of spreading may be related to one, two or all of the three mechanisms mentioned above. Currently available data prevent us to measure the precise effect of each of these mechanisms. However, in regard to these data, the second mechanism mentioned above, i.e., the combined effects of the rapid PSP-EU convergence and the “sea-anchor” force provided by the vertical position of the subducting Pacific slab at the Mariana Trench, likely provides a non-negligible driving force for asymmetric spreading in the Mariana Trough.
 We thank the crew of the R/V Hakuho-Maru and Kairei, and the KH92-1 and KR98-12 scientific parties. Fernando Martinez, Susumu Umino, and an anonymous reviewer are warmly acknowledged for useful reviews. This work is supported by JAMSTEC, Deep Sea Research Department.