Reconstructing deformation along the northwestern margin of the Tibetan Plateau is critical for evaluating the relative importance of microplate versus continuum models of the Indo-Asian collision. Questions regarding this margin's evolution are as follows: (1) What is the total offset along the sinistral Altyn Tagh strike-slip system? (2) How has that offset been absorbed in the western Kunlun Shan? (3) Why does the N-S width of the plateau vary along strike? Ion microprobe U-Pb zircon apparent ages of 17 plutons from NW Tibet, together with regional geologic observations, define a discrete, E-W trending boundary between two tectonic belts that has been offset along the Altyn Tagh system by 475 ± 70 km. Kinematic arguments indicate that this offset cannot be the result of north directed thrusting in the western Kunlun Shan. Therefore we propose that south directed faulting in the Tianshuihai thrust belt both offset the tectonic boundary and produced the asymmetry in the plateau. Shortening appears to have been absorbed in the upper crust by thin-skinned thrusting, in the middle/lower crust by east directed ductile flow and/or subduction, and in the mantle by north dipping subduction. Factors controlling the formation of the south directed thrust system appear to be the contrast between the rigid Tarim and the weaker Songpan-Ganzi flysch belt and a mantle suture inherited from late Paleozoic subduction. The evolution of western Tibet leads to a view of continental deformation that integrates elements of the microplate model (e.g., plate-like mantle and crust-mantle decoupling) with aspects of the continuum model (weak crustal flow beneath the plateau).
 Understanding the evolution of the Altyn Tagh-western Kunlun fault system is critical for distinguishing between the microplate and continuum explanations for the asymmetry of Tibet and thus for evaluating the relative significance of these two end-member views. Although the potential significance of the Altyn Tagh system is clear [Molnar and Tapponnier, 1978, 1975; Tapponnier and Molnar, 1976, 1977] two critical aspects of its evolution remain poorly understood: (1) the total strike-slip offset and (2) how this offset has been absorbed at the southwestern end of the fault system.
 We address these questions by measuring U-Pb zircon apparent ages to determine if batholithic belts in the western and eastern Kunlun Shan correlate (Figure 1c). These data constrain the position of a discrete tectonic boundary that has been offset along the Altyn Tagh system by 475 ± 70 km. As we demonstrate below, presently active north directed thrusts in the western Kunlun Shan cannot account for this offset. Therefore we hypothesize that slip was absorbed at the southwestern end of the Altyn Tagh fault by the Tianshuihai belt, a south directed thrust system that shortened western Tibet and produced the strong asymmetry in the north-south width of the Tibetan Plateau.
 U-Pb apparent ages were determined from 17 zircon samples using the UCLA CAMECA ims 1270 ion microprobe following procedures reported by Quidelleur et al. . Analyses were made using a 3–8 nA O− primary beam focused to a spot of ∼20 μm diameter. Secondary ionization of Pb+ was enhanced by flooding the sample with O2− at a pressure of ∼3 × 10−5 torr (1 torr = 133.3 N/m2). A mass resolving power of ∼6000 was used to distinguish the 204Pb peak from 176Hf28Si+, the principal molecular interference in zircon analysis [Compston et al., 1984]. The 238U+ peak was measured with a 3.5 or 5 eV offset. On average, ∼10 zircon grains were analyzed per sample, typically with a single spot analysis per grain. Weighted mean ages of zircon standard AS3 yielded a dispersion of <1% from the canonical value of 1099.1 ± 0.5 Ma determined by thermal ionization mass spectrometry [Paces and Miller, 1993].
 Isotope ratios and apparent ages for all samples are presented in Table 1, corrected for common Pb using the measured 204Pb and the weighted mean isotopic composition of modern aerosol lead in the Los Angeles basin [Sañudo-Wilhelmy and Flegal, 1994] (see Table 1). This composition was chosen based on the assumption that a significant fraction of the common Pb in the samples results from surface contamination introduced during sample preparation. To evaluate this assumption, we have also corrected the Pb/U ratios from the least radiogenic sample (9: AY94-9-12-3) using compositions for 0 and 450 Ma terrestrial Pb as calculated from a model of terrestrial Pb evolution [Stacey and Kramers, 1975]. As Table 2 indicates, 206Pb*/238U apparent ages calculated using the three common Pb compositions differ by less than 1%. Over 88% of the grains analyzed in the present study contain >95% radiogenic 206Pb (206Pb*) and are thus relatively insensitive to uncertainties in the common Pb correction. Application of the 204Pb correction (aerosol composition) to sample 14 and one set of analyses of grains from sample 13 overcorrects 207Pb, resulting in reversely discordant U-Pb ages. For these samples, grains with Th/U ≤0.5 were corrected assuming concordancy between the U-Pb and Th-Pb systems. Grains with Th/U >0.5 were corrected using 204Pb.
Table 1. Ion Microprobe U-Pb Analytical Results Corrected for Common Pb
Corrected using modern Los Angeles environmental Pb: 206Pb/204Pb, 18.911; 207Pb/204Pb, 15.70; 208Pb/204Pb, 38.526.
Corrected using zero-age Stacey-Kramers model Pb: 206Pb/204Pb, 18.703; 207Pb/204Pb, 15.529; 208Pb/204Pb, 37.859.
Corrected using 450 Ma Stacey-Kramers model Pb: 206Pb/204Pb, 17.99; 207Pb/204Pb, 15.589; 208Pb/204Pb, 37.116.
Uncorrected for common Pb.
Percent radiogenic 206Pb.
 Common Pb-corrected analyses typically define a cluster of concordant ages which span 50–70 Myr (Table 1). The distribution of 206Pb*/238U ages is shown in Figure 2 as histograms and cumulative probability plots. Six of the samples (2, 8, 9, 10, 13, and 14) contain older and/or younger analyses which are also concordant, but which were excluded from calculations of the weighted mean on the basis of the following criteria. All older analyses with peaks in the cumulative probability curve which are distinct from the maximum are interpreted as inherited zircons. Although samples 4, 7, and 11 show asymmetric 206Pb*/238U age distributions with old tails (Figure 2), possibly reflecting either protracted crystallization resulting from magmatic recharge or recycling of zircon from earlier intrusions, these analyses are included in the weighted mean age calculation. Evaluating the significance of the younger grains is more problematic. On average, the young grains in samples 2, 8, and 14 are 4, 14, and 46%, respectively, less radiogenic in 206Pb than the average of all other analyses from the same sample (Table 1). Because these young ages may result from either problems with the common Pb correction or analysis of grains which were more susceptible to younger Pb loss, these analyses were excluded from the calculation of the weighted mean. Because sample 13 is penetratively foliated, the anomalously young age from this sample was excluded due to the possibility that it reflects Pb loss during a younger metamorphic event. Excluded analyses are indicated in Table 3.
 Even when these outlying ages are not included in the weighted mean, mean square weighted deviation (MSWD) values are higher than is expected for single age populations, suggesting that either the assigned analytical errors are too small or the plutons contain multiple age populations. True variability of the single grain ages could be the result of entrainment of restitic zircons during melting in the source area, incorporation of xenocrystic zircon during emplacement, protracted crystallization due to slow cooling or pulsed intrusion, or variable amounts of Pb loss subsequent to crystallization. It must be emphasized that the goal of this study is to characterize the spatial and temporal distribution of magmatism in the western Kunlun Shan, Altyn Tagh, and Nan Shan areas, rather than to resolve detailed crystallization histories. It is clear from the data presented here that northern Tibet is characterized by regionally extensive early to middle Paleozoic magmatism.
3.3. Spatial Patterns in Ages
 U-Pb zircon and Rb-Sr ages from northwestern Tibet are compiled in Figure 3 and show two age populations: an older suite that ranges from early to middle Paleozoic (518–384 Ma) and a younger suite that ranges from late Paleozoic to early Mesozoic (290–180 Ma). In the following section, we separate samples in the western Kunlun Shan (i.e., between 76°E and 84°E longitude) from those in the eastern Kunlun Shan, Altyn Tagh, and Nan Shan ranges (i.e., east of 84°E). We refer to these as the western and eastern areas, respectively. Both areas contain the two age populations.
 In the western area, the older suite ranges from Early Ordovician (478 ± 9 Ma, sample 2, 1 standard deviation) to Middle Devonian (384 ± 2 Ma, sample E) (timescale from the work of Haq and Van Eysinga ). By contrast, in the eastern region, the older suite ranges from Middle Cambrian (518 ± 13 Ma, sample 7) to Middle Devonian (389 ± 5 Ma, sample 11). Despite these broad ranges, in both areas the older suite shows a pronounced concentration of Early to Late Ordovician ages (i.e., 450–475 Ma) (Figure 3).
 The younger population of samples spans an age range that is similar to the older suite. Samples A (277 ± 6 Ma) and H (180 ± 10 Ma) from the western area and samples 16 (290 ± 4 Ma) and Q (194 ± 17 Ma) from the eastern region indicate that the younger suite ranges from Early Permian to Early Jurassic. In the western area the younger suite shows a concentration of Early Jurassic ages between 180 and 215 Ma, whereas the eastern area has both an Early Jurassic concentration and a Late Permian to Middle Triassic peak.
 Although the temporal distributions of the two age populations are similar in both the western and eastern areas, the populations show important spatial variations from north to south. Figure 3 shows that samples from the older suite are restricted to areas north of the Karakax and Kunlun faults. The older plutons are widely distributed in these northern areas and show no systematic north-south variations in age, although such variations may be obscured by younger deformation or incomplete sampling. The younger suite of plutons is present both north and south of the Karakax and Kunlun faults. Thus in both the western and eastern regions, a mixed distribution of young and old plutons characterizes areas in the north while only the younger phase of plutonism is present to the south.
 The second belt is defined by the younger suite of plutons in Figure 3 and represents a Late Permian to Early Jurassic magmatic arc. This younger belt was locally superimposed upon the southern margin of the older belt, consistent with the interpretation that the younger belt formed during north directed subduction of the Paleotethys oceanic lithosphere [Matte et al., 1996; Mattern et al., 1996; Pan, 1996b; Sengör, 1984].
4.2. The Offset Marker
 Combining the timing of plutonism (Figure 3) with thermochronologic and regional geologic data (Figure 4) indicates that the matching plutonic belts coincide with a pair of tectonic provinces (Figure 5). As we show, recognition of these tectonic provinces is significant because they are separated by a sharp, east-west striking boundary that has been offset by 475 ± 70 km along the Altyn Tagh fault (Figure 5). This offset determination is similar to those proposed by Ritts and Biffi  and Chen et al. .
 Although a comprehensive discussion of the geology of the western and eastern Kunlun Shan is beyond the scope of this paper, essential aspects of these two areas are summarized in Figures 4a and 4b. The defining characteristics of the northern tectonic belt, the tectonic boundary, and the southern tectonic belt are listed in Figure 4b from top to bottom, respectively. Geologic relations from the western and eastern Kunlun Shan are also given.
4.2.1. Northern Tectonic Belt
Figure 4 indicates that the northern tectonic belt is an early to middle Paleozoic orogenic and plutonic complex that formed along the southern margin of the Tarim-North China craton. This early phase of plutonism was followed by a magmatically quiet period during which basement rocks were first exhumed during the development of regional unconformities and then buried by dominantly terrestrial deposits.
4.2.2. Southern Tectonic Belt
 In contrast to the northern belts, Figure 4 indicates that regions to the south are dominated by low-grade, Permo-Triassic slate belts of the Songpan-Ganzi complex that accreted to Eurasia during closure of the Paleotethys ocean basin [Sengör, 1984; Sengör et al., 1996; Sengör and Okurogullari, 1991]. These monotonous rocks are penetratively deformed by south vergent folds and are intruded by shallow level plutons [Matte et al., 1996]. Several recent papers [Xiao et al., 2002a, 2002b] have investigated the general tectonic setting of this belt in the western Kunlun Shan region.
4.2.3. Tectonic Boundary
 The boundary between the northern and southern tectonic belts is presently defined by the Karakax and Kunlun faults in the western and eastern Kunlun Shan, respectively. Both faults are active, left-lateral strike-slip faults that appear to have been superimposed upon an older suture zone, the Anyemaqen-Maqi or Kunlun suture [Dewey et al., 1988]. This tectonic boundary also represents an abrupt contrast in metamorphic grade, with higher-grade rocks intruded by the Kunlun batholith exposed to the north.
 It is important to accurately locate this boundary because it is a discrete regional marker that can be used to determine total offset along the Altyn Tagh fault system. The tectonic map in Figure 5 shows the location of the tectonic boundary as interpreted from the regional geology summarized in Figure 4 and the southernmost extent of samples from the older intrusive suite. In the eastern Kunlun Shan, samples 6 and 17 (Figure 3) provide a critical constraint on the position of the boundary on the southeast side of the Altyn Tagh fault system. These samples lie within 10 km of the active trace of the Altyn Tagh fault and indicate that the boundary between the northern and southern tectonic belts must lie between them. In the western Kunlun Shan, to the northwest of the Altyn Tagh system, the boundary lies southwest of samples L, M, and 4 and is bracketed by samples J and K. The arrowheads in Figure 5 indicate the tectonic boundary has been offset 475 ± 70 km along the Altyn Tagh fault.
 Do the present exposures in the western Kunlun Shan truly reflect the position of the tectonic boundary between the Paleozoic belt in the north and the Songpan-Ganzi flysch belt to the south? Although the following two scenarios are possible, in both cases the 475 ± 70 km offset is a minimum since the tectonic boundary would have originally been located south of its present position. One scenario is that the Songpan-Ganzi complex extends to the north beneath the Paleozoic belt. Because the Paleozoic belt is generally older than the Songpan-Ganzi complex, this older-on-younger configuration would require that the contact between the two belts is a north dipping thrust fault. In this case the southern edge of the hanging wall was originally located at, or to the south, of the present trace of the tectonic boundary. A second scenario is that the Paleozoic belt extends to the south, under the Songpan-Ganzi complex. In this case, the flysch belt was either obducted or deposited onto the southern edge of the Paleozoic basement. This scenario is unlikely since Permo-Triassic strata exposed within the Hotan thrust belt along the northern edge of the western Kunlun Shan (Figure 4) [Cowgill, 2001] do not correlate with the Songpan-Ganzi complex, indicating that such onlap cannot extend very far north of the present boundary. Also, the exposed boundary in the western Kunlun Shan appears to be a suture zone [Deng, 1996; Pan, 1996b; Yin and Harrison, 2000].
 Several factors contribute to the ±70 km uncertainty in the proposed offset. The largest contribution results from ambiguity in the location of the intersections between the Altyn Tagh fault system and the tectonic boundary. North of the fault, in the western Kunlun Shan, the boundary curves into the fault (Figure 5). Depending on how the boundary is projected along strike, the location of the intersection varies by 45 km. South of the fault, in the eastern Kunlun Shan, the intersection lies between samples 6 and 17 (Figure 5). Although these samples lie ∼150 km apart, field mapping [Cowgill, 2001] suggests that the pluton sampled at 17 continues ∼50 km east, reducing the interval to ∼100 km (Figure 5). The ±70 km uncertainty in our total offset determination results from the combination of the ±45 and ±100 km uncertainties.
 A second source of uncertainty is the planar nature of the offset boundary. Although intersection of the boundary and the Altyn Tagh fault forms a piercing line, rather than a piercing point, two arguments suggest that this uncertainty is not significant. First, both the fault [Cowgill, 2001] and the boundary [Matte et al., 1996; Mattern et al., 1996] dip steeply. The piercing line is thus nearly vertical. Second, 40Ar/39Ar analyses [Arnaud, 1992; Cowgill, 2001] suggest that broadly similar crustal levels are exposed adjacent to the fault in the western and eastern Kunlun Shan. We estimate that the error introduced by correlation of a piercing line is less than ∼50 km.
 A third contribution to the uncertainty results from the possibility that offset of the tectonic boundary was produced by a transform segment in the mid-Paleozoic arc, Mesozoic slip along the Altyn Tagh system [e.g., Delville et al., 2001; Sobel et al., 2001], or oroclinal bending of the tectonic boundary. Although it is difficult to quantify these uncertainties, we argue that such errors can be neglected because these scenarios are unlikely. It is unlikely that separation reflects an original transform segment because deformation at the southwest end of the Altyn Tagh system appears to postdate Mesozoic accretion of the Qiangtang and Lhasa blocks. In addition, offset of a Tertiary thrust belt by ∼380 km along the eastern Altyn Tagh fault [Yin and Harrison, 2000] indicates that a significant fraction of the 475 ± 70 km total offset is Cenozoic. Finally, it is unlikely that the separation between the western and eastern Kunlun Shan results from oroclinal bending, since the tectonic boundary approaches the fault system at a high angle.
 As Figure 6a indicates, the problem with the present structural configuration is that it places both the western and eastern Kunlun Shan belts in the same plate. As Figure 6b indicates, displacement of the tectonic boundary requires that the western and eastern Kunlun Shan batholiths were originally located within separate plates so that the eastern Kunlun Shan and Tibet could move to the east-northeast relative to the western Kunlun Shan and Tarim. Thus offset of the western and eastern Kunlun Shan requires that a major fault system lies to the south of the western Kunlun batholith.
 One possibility is that this structure is a strike-slip fault. In this case, offset of the tectonic boundary is due to eastward extrusion of the Tibetan Plateau via simultaneous slip on the conjugate Karakorum and Altyn Tagh/Karakax faults [Molnar and Tapponnier, 1978; Peltzer and Tapponnier, 1988; Tapponnier et al., 1982]. However, there are three problems with this model. First, the total offset along the southern Karakorum fault ranges from ∼120 km [Searle, 1996] to only ∼60 km [Murphy et al., 2000], limiting the extent to which this fault could have accommodated enough eastward extrusion to create the asymmetry of the Tibetan Plateau. Second, as Figure 7 indicates, the extrusion model predicts ∼35° of clockwise, vertical axis rotation of the western Kunlun Shan and Karakax fault. Paleomagnetic directions measured from Cretaceous and early Tertiary strata north and south of the western Kunlun batholith (Table 4 and Figure 5) [Chen et al., 1992, 1993; Gilder et al., 1996; Rumelhart et al., 1999; Yin et al., 2000] are inconsistent with such rotation. Because rocks to the north and south of the western Kunlun belt have not been appreciably rotated, it is likely that the intervening Kunlun belt also has not been rotated. Finally, because the western and eastern Kunlun Shan have similar strikes, it is simpler to conclude that they were initially linear, rather than initially kinked (e.g., Figures 7b and 7c) and subsequently straightened by vertical axis rotation of the western Kunlun Shan (e.g., Figures 7d and 7e).
Table 4. Compiled Paleomagnetic Data From Southern Tarim and Western Tibet
 To explain the offset of the tectonic boundary, we propose that left slip motion along the Altyn Tagh fault system was primarily accommodated at the southwestern end of the fault by the Tianshuihai thrust system, a south directed thrust belt that we suggest lies to the south of the western Kunlun batholith (Figures 6b and 8). We refer to this system as a back thrust belt because deformation within this thrust belt was directed into the interior of the Tibetan orogen. As discussed below, there are two essential aspects to the proposed back thrust belt. First, we argue that the style of shortening within the belt varied with depth. In particular, we propose that (1) the upper crust has been imbricated by a south directed fold-and-thrust belt, (2) the middle/lower crust has been removed either by east directed horizontal flow or subduction into the mantle, and (3) the mantle lithosphere has subducted northward beneath Tarim. Second, we suggest that the primary control on the location, geometry, and style of deformation within the hypothesized thrust system was caused by the contrast in strength between the strong lithosphere of Tarim and the weaker crust of the Songpan-Ganzi flysch complex, in conjunction with a relict, north dipping mantle suture inherited from the Paleozoic-Mesozoic assembly of Tibet.
 An important aspect of the proposed back thrust is that it explains the marked westward decrease in the north-south width of the Tibetan Plateau at the southwestern end of the Altyn Tagh fault. The distance between the Indus-Yalu suture in the south and the Kunlun/Anyemaqen-Maqi suture to the north is ∼750 km near 90°E longitude but only ∼250 km near 80°E (Figure 1b). The 500 km decrease in the north-south extent of the Plateau is equivalent to the 475 ± 70 km total left-lateral offset we are proposing along the Altyn Tagh fault system. Assuming this westward narrowing of the Tibetan Plateau does not reflect its initial geometry, it must have been produced by structures that lie between these two geologic markers. The westward narrowing of the Tibetan Plateau is mirrored at the northeastern end of the Altyn Tagh system by an eastward decrease in the width of the Tarim Basin into the Hexi Corridor (Figures 1a and 1c). We suggest that the along-strike asymmetry of the Tibetan Plateau and Tarim Basin developed because left slip on the Altyn Tagh system was accommodated at the western and eastern ends of the fault by a pair of thrust belts with opposite polarity, with northeast directed thrusting in the Nan Shan and Qilian Shan thrust belt and southwest directed thrusting within the Tianshuihai thrust belt.
 Several observations support the proposed south directed thrusting within western Tibet:
 2. As previously mentioned, the Karakax fault juxtaposes higher-grade rocks to the north against lower-grade units to the south, suggesting that the western Kunlun Shan has been uplifted relative to Tibet [Matte et al., 1996]. K-feldspar 40Ar/39Ar age spectra from samples to the north of the Karakax fault show age gradients extending into the early Tertiary [Arnaud, 1992], possibly reflecting thermal perturbation of these samples in the Cenozoic.
 3. North dipping foliations and south vergent structures are present in the hanging wall of the Karakax fault [Norin, 1946] and the Songpan-Ganzi complex to the south [Matte et al., 1996]. Near Shengli (locality “d” on Figure 4) Middle Jurassic volcanic rocks are thrust southward over tightly folded Jurassic terrestrial deposits [Matte et al., 1996] indicating that this deformation is younger than Middle Jurassic. Within the Songpan-Ganzi flysch complex, some of the folds between Mazar and Kangxiwar (Figure 4) are south vergent, as are isoclinal folds in the vicinity of Aksai Chin (Figure 5) [Matte et al., 1996]. Aureoles around Jurassic plutons locally overprint cleavage within the slate belt [Matte et al., 1996] indicating that at least some of this deformation is pre-Tertiary in age. However, two phases of folding are evident in the Bazar Dara slates [Gaetani et al., 1990]. We suspect that the extent and magnitude of the younger phase of deformation have not been fully appreciated due to the paucity of marker beds and the south vergent nature of both phases of deformation.
 In Figure 9 we have reconstructed the Cenozoic evolution of the Altyn Tagh system and its adjacent thrust belts to illustrate how this system might have evolved. This reconstruction illustrates that offset estimates for individual structures within the Altyn Tagh-western Kunlun system can be combined to balance deformation along the northern margin of the Tibetan Plateau.
 To reconcile early, south directed thrusting in the western Kunlun Shan region with the present north directed deformation, we suggest that Cenozoic left slip along the Altyn Tagh fault has been accommodated by two stages of deformation (Figure 9). In the first phase, deformation at the southwestern end of the fault system was dominated by south directed thrusting in the Tianshuihai thrust belt. In the second stage, deformation switched to dominantly north directed thrusting along the northern margin of the range. The extent to which these thrust systems were active simultaneously is unclear. One possibility is that cessation of south directed thrusting in the Tianshuihai belt coincided with the initiation of north directed thrusting along the Hotan and Tiklik faults (Figure 4). However, it is more likely that both thrust systems were active simultaneously and that a drop in the shortening rate of the Tianshuihai belt coincided with an increase in the shortening rate to the north. In either case, we suspect that the reversal in thrust polarity was triggered by uplift of the Tibetan Plateau.
 From the Eocene to the late Miocene (Figure 9), we infer that left slip on the Altyn Tagh fault was accommodated at the western end of the fault primarily by the Karakax fault and north dipping thrusts within the Tianshuihai thrust belt to the south. In contrast, slip at the eastern end of the fault was accommodated by south dipping thrusts within the Nan Shan and Qilian Shan. As a result of this geometry, the eastern Kunlun Shan was coupled with Tibet and both moved northeast relative to the western Kunlun Shan that formed part of Tarim. Separation between the western and eastern Kunlun Shan accumulated while this structural configuration was active (see Figure 6b).
 Between Oligocene to late Miocene, the Tula-Subei segment of the Altyn Tagh fault formed (Figure 9), cutting the Jinyang/Suekuli-Danghe/Yema Nan Shan thrust belt and initiating left separation of the two halves of this Tertiary thrust belt [Gehrels et al., 1999; Yin and Harrison, 2000]. Slip along the North Altyn fault either fed back into the Tula-Subei segment of the Altyn Tagh fault as part of a rhomb-shaped transpressional duplex [Cowgill et al., 2000] or it continued into Mongolia, linking with left slip faults described by Lamb et al. .
 North directed thrusting along the northern edge of the western Kunlun Shan had initiated at or prior to ∼37 Ma [Rumelhart, 1998]. We interpret this thrust system to have initiated as a conjugate to the main, south directed Tianshuihai thrust belt. In this interpretation, most of the Oligo-Miocene relative motion between Tarim and Tibet was accommodated by shortening within the Tianshuihai thrust belt. As deformation progressed, it appears that north directed faulting within the western Kunlun thrust belt accelerated at the expense of shortening within the Tianshuihai thrust belt (Figure 9). This kinematic reorganization of the west Kunlun-Altyn Tagh structural system transferred the western Kunlun Shan from the Tarim block to the Tibetan Plateau, thereby terminating large-magnitude relative motion between the western and eastern Kunlun Shan (Figures 6 and 9).
 We estimate the total offset along the Minfung-Qiemo section of the western Altyn Tagh fault (Figure 5) to be ∼575 km. This value combines the Tibet-Tarim offset that was absorbed by ∼100 km of north directed thrusting along the Hotan and Tiklik faults (Figure 4) [Cowgill, 2001; Lyon-Caen and Molnar, 1984] and 475 ± 70 km of offset of the tectonic boundary. East of Qiemo (Figure 5), structural mapping indicates that rocks on opposite sides of the North Altyn fault do not match for ∼120 km along strike, suggesting that left slip separation is at least this large [Cowgill et al., 2000]. Some of this strike-slip deformation may have been responsible for the initial development of the Jinyang/Suekuli-Danghe/Yema Nan Shan thrust belt. This thrust belt has subsequently been dissected and offset by an additional 280 ± 30 km along the Tula-Subei segment of the eastern Altyn Tagh fault (Figure 5) [Yin and Harrison, 2000]. Total offset on the Qiemo-Subei section of the eastern Altyn Tagh fault system is thus ∼400 km, equivalent to the offset determined from correlation of a Jurassic facies boundary across the fault [Ritts and Biffi, 2000]. In our reconstruction (Figure 9), 180 km of shortening within the Qiman Tagh accounts for the eastward decrease in offset between the Minfung-Qiemo and Qiemo-Subei segments of the Altyn Tagh fault system.
 Previous work in the North Pamir and Paropamisus blocks (Figure 1c) [e.g., Boulin, 1988; Girardeau et al., 1989; Stöcklin, 1989] suggests that the tectonic boundary continues west of the Kunlun Shan [Burtman and Molnar, 1993; Sengör, 1984]. The boundary appears to coincide with the Akbaytal and Herat faults in the North Pamirs and Afghanistan, respectively (Figure 1c). By the late Miocene, indentation of the Pamir syntaxis had begun to deform the western section of the Kunlun-Tianshuihai belt (Figure 9). Southward subduction of the Tarim-Tadjik crust beneath the Pamirs, as well as internal shortening within the syntaxis displaced the tectonic boundary 300+ km to the north [Burtman and Molnar, 1993]. We estimate that the boundary shows ∼360 km of sinistral offset and ∼270 km of dextral offset along the western and eastern sides of the Pamirs, respectively (Figure 9). The mapped pattern of deformation suggests that oroclinal bending has accommodated a significant fraction of this displacement with the remainder occurring as slip along the Chaman and Karakorum fault systems (Figure 9). Total offset on the east side of the syntaxis is ∼90 km less than on the west side because the west Kunlun thrust belt has moved north relative to both the Tarim and Paropamisus along the north directed Kunlun thrust system.
 It is possible to calculate the long-term slip rate of the Altyn Tagh fault system by combining the total offset on the Altyn Tagh system with estimates of its initiation age. Several lines of evidence indicate that the Altyn Tagh fault system had begun to develop by at least 37 ± 5 Ma, including magnetostratigraphic constraints on the timing of the onset of coarse-clastic deposition within the Tarim Basin [Rumelhart, 1998; Yin et al., 2002], apatite fission track data from the western Kunlun Shan and Altyn Tagh [Cowgill, 2001; Rumelhart, 1998; Sobel, 1995; Sobel et al., 2001; Yin et al., 2002], and the timing of reorganization of sediment dispersal patterns within the northwestern Qaidam basin [Hanson, 1997]. A mid-Oligocene initiation age for the central Altyn Tagh fault is also consistent with cooling histories derived from interpreting 40Ar/39Ar K-feldspar results in the context of multidiffusion domain theory [Cowgill et al., 2001].
 By combining this initiation age with our total offset estimate of 475 ± 70 km we calculate a 37 Ma average slip rate of 13 ± 3 mm/yr. This rate is equivalent, within error, to rates determined from a recent global positioning system (GPS) study (9 ± 2 mm/yr) [Shen et al., 2001], and is consistent with results from a recent paleoseismic study of the central Altyn Tagh fault [Washburn et al., 2001]. As discussed above, the total offset across the western Altyn Tagh fault system (between Minfung and Qiemo in Figure 5) is estimated to be ∼575 km, in which case the long-term slip rate of the system as a whole is 16 ± 3 mm/yr.
6.1. Structural Style as a Function of Depth
 If the Tianshuiahi back thrust hypothesis is correct, then the following mass balance indicates that a significant volume of crust has been removed from western Tibet during Cenozoic shortening. According to the back thrust hypothesis, the Tianshuihai thrust belt has accumulated ∼475 km of shortening. Since this belt is presently ∼200 km wide, its original width must have been ∼675 km indicating that the belt has been shortened by ∼68%. Because the present crustal thickness of the inferred thrust belt is ∼63 km [Ma, 1980], northeast-southwest sections across the belt have an area of ∼12,600 km2. The original crustal thickness can be calculated using the observation that widespread limestones near Aksai Chin and the Ghoza-Longmu Co (Figure 5) [Matte et al., 1996; Norin, 1946] were deposited near sea level between the Aptian and Cenomanian [Matte et al., 1996]. By assuming that these rocks were uplifted to their present elevation of 5000+ m by Airy isostatic compensation of regionally thickened crust, we calculate [e.g., Turcotte and Schubert, 1982, equation 5.146] that the crust was ∼30 km thick when they were deposited. In this case, northeast-southwest cross sections across the 675 km wide belt had an initial area of ∼20,250 km2. Comparison of initial and final sections indicates ∼40% of the original crustal thickness has been lost from western Tibet.
 Although the simplest explanation is that the excess mass has been removed by erosion, this is unlikely since it would produce widespread exposures of midcrustal rocks in western Tibet, rather than the observed slates of the Tianshuihai belt. A second possibility is that the excess crust has been removed by the subduction of either remnant oceanic lithosphere or dry lower continental crust. As described below, the Songpan-Ganzi complex accreted to the southern margin of Eurasia during doubly vergent subduction of the Paleotethys ocean basin [e.g., Sengör, 1984; Sengör and Natal'in, 1996]. If Mesozoic subduction was incomplete, then prior to the onset of the Indo-Asian collision the Tianshuihai complex would have been underlain by a relict oceanic slab that was trapped between the Lhasa-Qiangtang blocks to the south and the Tarim-North China blocks to the north. The Black and Caspian Seas are essentially modern analogues for such a setting, although Cretaceous shallow marine deposits in western Tibet suggest that the Paleotethys basin had been filled with sediment prior to the Indo-Asian collision. Reactivation of subduction during Cenozoic shortening could then remove the oceanic crust that floored this basin. The lack of extensive subduction-related volcanics in southern Tarim and western Tibet requires a mechanism for rejuvenating subduction without producing an associated volcanic arc. However, if Mesozoic subduction completely removed the oceanic lithosphere during hidden subduction [i.e., Sengör, 1984], then the asthenosphere could have been emplaced against the base of the accretionary complex, causing it to undergo widespread dehydration. Dry, metasedimentary xenoliths of the lower crust in central Tibet [Hacker et al., 2000a] may have formed by such a process. Subduction of dry lower crust may account for the loss of mass from western Tibet without the construction of a magmatic arc. Although in these subduction scenarios the Tianshuihai complex is underlain by either relict oceanic crust or dehydrated metasediments, another likely scenario is that most of the crust consists of the same hydrous, quartz-dominated lithologies exposed in western Tibet, and a similar crustal structure has been proposed for central Tibet [Kapp et al., 2000]. According to the compilation of flow laws reported by Carter and Tsenn , crust that is dominated by wet, quartz-rich lithologies will have minimal strength (<10 MPa) below 12 km depth for a typical continental geotherm and a strain rate of 10−14 s−1 [see also Brace and Kohlstedt, 1980; Molnar, 1992; Ranalli and Murphy, 1987]. Thus a third possibility is that the excess crust of western Tibet was removed by ductile flow of a weak middle/lower crust. Due to the narrow north-south width of the Pamirs (Figure 1), this crust may have been expelled eastward, helping to inflate the central Tibetan Plateau (Figure 8b). Weak middle to lower crust beneath the Tibetan Plateau has also been inferred from geophysical studies [Nelson et al., 1996 and references therein] and geodynamic models that incorporate a depth-dependent viscosity [Clark and Royden, 2000; Royden, 1996; Royden et al., 1997; Shen et al., 2001]. To balance the ∼500 km of crustal shortening at depth, we suggest that the “Tibetan” (i.e., Qiangtang and Songpan-Ganzi) mantle lid was subducted northward beneath Tarim (Figure 8b).
6.2. Why Did the Tianshuihai Back Thrust System Form?
 Above we argue that Cenozoic left slip along the Altyn Tagh fault system has been absorbed at the southwestern end of the fault by thin-skinned, south directed thrusting within the Tianshuihai slate belt. Structures inherited from the Paleozoic and Mesozoic assembly of Eurasia appear to have been important in controlling the development of this system.
 Two aspects of this history appear to be particularly significant with respect to the development of the Tianshuihai back thrust system. First, we suggest that termination of north directed subduction in the Triassic produced a north dipping suture within the mantle lithosphere beneath the western Kunlun Shan that reactivated during the Indo-Asian collision. It is likely that the upper plate of this mantle suture zone consisted of hydrated mantle, while the suture itself could have been coated with serpentinites and possibly metasedimentary units that had been plated onto the walls of the subduction thrust. If these materials were present, then it is possible that the mantle suture could have remained a plane of weakness following closure of the Paleotethys ocean basin. If such a relict suture has been reactivated within the western Kunlun Shan, then it suggests such sutures may remain weak for more than 200 Myr following their formation. The longevity of such a weakness may result from the compositional changes produced by hydration of the upper plate mantle during prolonged subduction of oceanic lithosphere. It is possible therefore that the longevity of such weak sutures may be controlled by the intensity of upper plate alteration, and thus the duration of subduction, that preceded formation of the suture.
 Second, during reactivation of the mantle suture it appears that the Tianshuihai back thrust system developed by propagation of deformation away from the strong Tarim lithosphere and into the weaker Songpan-Ganzi flysch belt. The continuity of Phanerozoic deposits within the Tarim Basin indicates that the basin has been little deformed since the late Paleozoic [Jia, 1997; Li et al., 1996; Sengör et al., 1996] and attests to the strength of the Tarim lithosphere. By contrast, much of the deeper crust of the central Tibetan Plateau may be composed of weak Songpan-Ganzi mélange that underplated the Qiangtang block during the Late Triassic [Kapp et al., 2000]. Kapp et al.  suggest that this mélange may be weaker and wetter than the surrounding continental basement, thus explaining the high crustal Poisson's ratio [Owens and Zandt, 1997] and widespread Cenozoic volcanism of northern Tibet [Deng, 1978], as well as the more intense Cenozoic shortening in northern Tibet [Coward et al., 1988] than in the Lhasa terrane to the south [Coward et al., 1988; Murphy et al., 1997]. Although deep crustal xenoliths are not consistent with present-day crustal melting in northern Tibet, they do indicate that the lower crust of Qiangtang is at least partly metasedimentary [Hacker et al., 2000a]. Geodynamic models incorporating a strong Tarim basin region also show asymmetric plateau development [Neil and Houseman, 1997; Soofi and King, 2002].
6.3. What Controls Continental Deformation?
 Our reconstruction of the tectonic evolution of western Tibet and the Altyn Tagh fault system suggests that vertical and lateral variations in the strength of the Eurasian continental lithosphere played a first-order role in determining where, when, and how Indo-Asian convergence was absorbed to form the Tibetan Plateau. Above we present three hypotheses to explain 475 ± 70 km of left separation between the western and eastern Kunlun Shan. First, we suggest that western Tibet has been shortened by the south directed Tianshuihai back thrust belt at the southwestern end of the Altyn Tagh fault system. Second, we argue that shortening within this belt was accommodated in the upper crust by thin-skinned thrusting, in the mantle lithosphere by north directed subduction, and in the middle/lower crust by either east directed flow or subduction of relict oceanic basement or dry lower crust. Third, we propose that the back thrust system formed by reactivation of a north dipping mantle suture and localization of deformation in the weak Songpan-Ganzi complex rather than the stronger Tarim.
 If the factors controlling the geometry and evolution of Cenozoic deformation in western Tibet generally hold, then they lead to a conceptual model of crustal deformation that integrates key aspects of both the microplate and continuum views. In particular, the tectonics of western Tibet suggest that two parameters are key in controlling how a continent deforms in response to a given set of applied boundary conditions (1) the existence, geometry, and strength of sutures within the mantle lithosphere and (2) lateral variations in the strength of the middle to lower crust. The synthesized view is one in which plate tectonics holds in the mantle (microplate theory), but the upper crust may or may not track motion of underlying plates, depending on strength of middle to lower crust (continuum theory). Where the middle/lower crust is weak, the upper crust is detached, potentially allowing the crust to thicken by crustal inflation. In such cases, the kinematics of deformation in the upper crustal and mantle lithosphere will be decoupled. Where the middle/lower crust is strong enough to keep the upper crust firmly coupled to the mantle lid, upper crustal deformation does track deformation in the lower crust and upper mantle. Thus lateral variations in preexisting strength of both the mantle and crust will play a profound role in controlling how continents deform.
 The following conclusions can be drawn from our investigation of the Altyn Tagh-western Kunlun fault system.
 1. U-Pb ion microbe analyses of zircons from 17 plutons from the northwestern margin of the Tibetan Plateau document a matching pair of plutonic belts in the western and eastern Kunlun Shan. A dominantly Ordovician belt is found only north of the Karakax and Kunlun faults while an Early Permian to Early Jurassic belt occurs both north and south of these structures.
 2. Combining the timing of plutonism determined from zircon geochronology with thermochronologic and regional geologic data indicates that the matching plutonic belts coincide with a pair of tectonic provinces. The northern tectonic belt is an early to middle Paleozoic orogenic and plutonic complex that formed along the southern margin of the Tarim-North China craton. The southern tectonic belt is characterized by the Songpan-Ganzi flysch belt that was accreted to this margin upon closure of the Paleotethys ocean. These belts are separated by a sharp east-west striking boundary that has been offset by 475 ± 70 km along the Altyn Tagh fault.
 3. Recent left-lateral slip along the Altyn Tagh system has been accommodated at the western end of the fault by north directed thrusting. However, it is unlikely that this structural system produced ∼475 km offset between the western and eastern Kunlun Shan since both belts lie within Tibet. To explain offset of the western and eastern Kunlun Shan we propose that north directed faulting at the eastern end of the Altyn Tagh fault was initially mirrored by south directed back thrusting at the west end of the fault along the Karakax fault and Tianshuihai thrust belt. Tibet-Tarim convergence across these belts resulted in the extreme narrowing of both the Tibetan Plateau in western Tibet and the Tarim basin in the Hexi corridor.
 4. Within the back thrust belt, upper crustal shortening within the hypothesized back thrust belt appears to have been accommodated by thin-skinned thrusting. To balance this shortening at depth we argue that the middle/lower crust of the Songpan-Ganzi complex and the Qiangtang block was either weak, and has been evacuated by east directed flow, or it was strong and was subducted into the mantle. One explanation for a strong lower crust is that it was a relict oceanic slab that floored a remnant ocean basin similar to the Black or Caspian Seas. Another possibility is that the lower crust was dry and strong due to metamorphism and/or partial melting extraction following foundering of the subducting slab during closure of the Paleothethys ocean basin. We suggest that shortening of the western Tibetan mantle lithosphere was accommodated by north dipping subduction.
 5. Structures inherited from the late Paleozoic to Mesozoic assembly of Eurasia appear to play a critical role in dictating the style of deformation at the western end of the Altyn Tagh system. Specifically, we propose that the back thrust system formed by reactivation of a north dipping mantle suture and localization of deformation in the weak Songpan-Ganzi complex rather than the stronger Tarim.
 6. A model of continental deformation, which integrates key aspects of the microplate and continuum views, appears to provide the most accurate description of continental deformation within the Indo-Asian collision. Specifically, the continental mantle lithosphere appears to be broken into microplates along lithospheric scale structures such as the Altyn Tagh fault [Tapponnier et al., 2001; Wittlinger et al., 1998]. However, the extent to which upper crustal deformation reflects the motion of these underlying blocks is a function of the strength of the middle to lower crust. In some regions, such as Tarim, the middle to lower crust is strong enough to maintain coupling between the brittle upper crust and the underlying block of mantle lithosphere; whereas elsewhere, such as within the Songpan-Ganzi complex in western Tibet, the upper crust has decoupled from the mantle lithosphere [Clark and Royden, 2000; Royden, 1996; Shen et al., 2001], allowing the weak middle to lower crust to evacuate from this area to thicken the crust of central Tibet. The mechanical behavior of continental crust thus appears to be a strong function of its geologic evolution.
 Eric Cowgill thanks Chris Coath and Kevin McKeegan for their instruction in the use of the CAMECA ims 1270 ion microprobe and their assistance during zircon analyses and subsequent data reduction. We thank Cindy Ebinger, George Gehrels, and Carole Petit-Mariani for their reviews. Earlier versions of the manuscript were significantly improved by reviews from Cindy Ebinger, Todd Ehlers, Laurent Jolivet, Paul Kapp, Nadine McQuarrie, Mike Murphy, Edward Sobel, and Alex Webb. Discussions with Alex Robinson, Yuan Chao, Kari Cooper, Mike Taylor, and Jorge Vazquez also significantly improved the clarity of the ideas presented here. We thank Laura Veirs, Jamie Buscher, and Elizabeth Catlos for field assistance. Fieldwork was made possible by collaborators Chen Zhengle, Zhang Qing, and Zhang Shuanghong, and drivers from the Tarim Petroleum Corporation. Cumulative probability plots in Figure 2 were generated using Isoplot/Ex v 2.3 provided by Kenneth Ludwig. This project was supported by NSF grants EAR9614664 and EAR9725599. We acknowledge facility support from the Instrumentation and Facilities Program of the National Science Foundation.