The seismic structure of the continent-ocean transition (COT) at magma-poor rifted margins can explain geological processes leading to continental breakup. At the Newfoundland-Iberia rift, compressional seismic velocity (Vp) is interpreted with multichannel seismic reflections and drilling results to document continental crustal stretching and thinning, exhumation of the mantle, and incipient seafloor-spreading. However, Vp cannot uniquely constrain COT geology. We present an updated 2-D model for Vp and a new shear-wave velocity model (Vs) for SCREECH Line 2 on the Newfoundland margin using multichannel seismic reflections and coincident ocean-bottom seismometer refraction data. In shallow COT basement we find Vp / Vs ratios average 1.77, which is normally too high for upper continental crust and too low for serpentinized mantle. This observation can be explained by stretching of a mafic middle and/or lower continental crust into the COT. We further support the presence of hydrated mantle peridotites at depth during rifting.
 Slow rifting at magma-poor margins often occurs without the weakening effects of magmatic diking. Continental crust may be therefore stretched by a large factor (β >5), and rocks from the lower continental crust and mantle may be brought to shallow depths via detachment faults [Whitmarsh et al., 2001]. During the final phases of rifting, brittle faulting and hydrothermal circulation can lead to serpentinization of mantle peridotites, weakening the lithospheric mantle [Manatschal, 2004; Reston and McDermott, 2011]. Before complete breakup of the lithosphere and the onset of normal seafloor-spreading, a ~100 km wide zone of continental mantle may thus be exhumed to the surface in continent-ocean transition zones (COT) [Whitmarsh et al., 2001]. The polarity of lithospheric detachment faults with respect to the locus of breakup often results in asymmetry at conjugate magma-poor margins [Manatschal, 2004; Lavier and Manatschal, 2006]. In the footwall of a detachment, small amounts of lower crust and, more extensively, uppermost mantle rocks are exhumed to form the COT. At the conjugate rift flank, continental crust of the hanging wall is often highly thinned and stretched above the upper mantle. Strain localization in weakened mantle rocks eventually leads to complete lithospheric breakup and a gradual increase in melt supply from the upwelling asthenosphere [Lavier and Manatschal, 2006].
 Continental rifting may be accompanied by small amounts of magmatism even if the lithospheric mantle is relatively cold or significantly depleted during rifting [Müntener and Manatschal, 2006]. Although rifting of thick continental lithosphere often produces ample magma by decompression melting, only limited amounts of extrusive volcanism may reach the surface if the extension rate is very low [Lizarralde et al., 2004]. Syn-rift melts can therefore be trapped below thick COT lithosphere, so it is possible that not all magmatism is accounted for in an ocean-bottom seismometer (OBS) refraction study of a rifted margin [Bronner et al., 2011]. A better understanding of the melting history of nominally magma-poor margins is important, because even a small amount of melt introduced in the lithosphere during rifting can alter the style of deformation [Kaczmarek and Müntener, 2010].
 The Newfoundland-Iberia conjugate margins are a prime example of mature magma-poor rift systems [Tucholke et al., 2007]. A large wealth of data from drilling expeditions and marine geophysical studies here document brittle extension of continental crust, exhumation and serpentinization of continental mantle, and a slow onset of seafloor-spreading. The distal Iberian margin is interpreted as almost entirely exhumed continental mantle with a wide zone of compressional seismic velocities (Vp) between 7.0 and 7.5 km/s [Whitmarsh et al., 2001]. This high-velocity zone in the Iberian COT is capped by lower velocities (Vp = 4–5 km/s) interpreted as mantle rock that was pervasively serpentinized after exhumation [Dean et al., 2000]. A similar evolution has been proposed for the conjugate margin in the Newfoundland Basin [Sibuet et al., 2007], although seismic images from the SCREECH project (Studies of Continental Rifting and Extension on the Eastern Canadian Shelf; Figure 1) suggest that crust in the COT is significantly different. Multichannel seismic (MCS) reflections [Shillington et al., 2006] demonstrate that the Newfoundland Basin lacks the faulted allochthonous crustal blocks that have been imaged off the Iberian margin [Krawczyk et al., 1996]. In addition, Vp in the shallow basement of the Newfoundland Basin is often lower than 6.0 km/s [Lau et al., 2006; Van Avendonk et al., 2006], which may indicate unroofed continental crust or exhumed, highly serpentinized mantle peridotites.
 Interpretation of the structure and evolution of the COT is complicated by strong seismic reflections from a package of postrift diabase sills just above the Newfoundland basement at SCREECH Line 2 [Shillington et al., 2008; Péron-Pinvidic et al., 2010]. A large injection of mafic material likely accompanied off-axis volcanism following the complete breakup of the Newfoundland-Iberia lithosphere and production of oceanic crust by the early Albian [Karner and Shillington, 2005; Tucholke et al., 2007]. We present an integrated analysis of travel-time constraints from MCS reflection and compressional OBS refraction and reflection data to better characterize the postrift sills and underlying basement on SCREECH Line 2. As a result, we are able to present a new analysis of shear waves that sample the Newfoundland COT basement, and we develop smooth seismic velocity models with regularized tomographic inversions of P- and S-wave travel times. We use both Vp and Vp / Vs ratio models in our geological interpretations of the rifted margin, assessing our interpretations by plotting seismic velocities with depth at key SCREECH Line 2 intervals. We also use shear-wave data to test simple bulk Vp / Vs models for the flat COT basement.
2 Travel-Time Tomography
 The travel times of reflected and refracted phases observed in marine seismic data can be inverted for a layered seismic velocity model with smoothly varying seismic velocities [Van Avendonk et al., 2004, and references therein]. Observation of a series of distinct seismic reflections and refractions in OBS data (Figure 2) provides evidence for layering near the top of crystalline basement that must be included explicitly in the seismic velocity model if we want to correctly model the travel times of later arrivals. We therefore express travel-time residuals for compressional seismic arrivals δTp,i as a sum of perturbations δVp and perturbations in the depth of layer boundaries δz [Thurber, 1985; Van Avendonk et al., 2004]. If ray paths calculated in the reference model are good estimates of the true source-receiver paths, this relationship can be approximated with a linearization
 The first term of (1) accounts for the contribution to δTp,i of seismic velocity perturbations δVp by integration over the ray path. The coefficient Γi,j represents the partial derivatives of travel-time δTp,i with respect to the depth of velocity boundary zj in the model [Stork and Clayton, 1991]. We use (1) to express the model constraints of wide-angle reflections and refractions in the OBS data of SCREECH Line 2 [Van Avendonk et al., 2006], and to incorporate two-way travel times from coincident MCS data [Shillington et al., 2006; Péron-Pinvidic et al., 2010].
 To find a smooth seismic velocity model we add spatial first- and second-derivative constraints on the structure of Vp and z to the system of equations ((1)), and solve them for a model perturbation in a least-squares inversion [Van Avendonk et al., 2004]. This linear inversion step is followed by raytracing in the updated seismic velocity model to calculate new ray paths and travel times. We iterate the raytracing and linearized inversion procedures until we obtain a smooth seismic velocity model that fits the travel-time data within a specified tolerance level.
 Once we have estimated Vp in a profile along SCREECH Line 2, we use a similar procedure to construct a Vs model from shear-wave travel times. However, unlike compressional wave arrivals, shear waves are not consistently observed in OBS records, and their travel times have relatively large uncertainties (100 ms) [Mjelde et al., 2003; Eccles et al., 2009]. We therefore cannot constrain the Vs structure along our profile with the same accuracy as the Vp image. We must also assume that the two models have seismic velocity discontinuities at the same depths zj. In most studies Vp and Vs appear well correlated [Brocher, 2005] such that their ratio R = Vp / Vs may be a more smoothly varying medium property than Vs itself. Accordingly, we express the shear-wave travel-time residual δTs,i in R, and we omit terms that assume a dependence on further perturbations of the model layer boundaries zj
 As in the case of the inversion of δTp,i, we add smoothness constraints for R and solve (2) for this parameter in a least-squares inversion.
3 P-Wave Modeling
 We use wide-angle [Van Avendonk et al., 2006] and near-vertical seismic travel-time constraints [Shillington et al., 2006] on the sediments and shallow basement structure along SCREECH Line 2 in a joint inversion of these data with the methodology described in the previous section. The new Vp model, which has detailed information on the sediment and basement interface structure, helps us understand the nature of P-S converted waves, which we must interpret before we invert shear-wave travel times for a model of Vp / Vs along SCREECH Line 2.
 The choice and geometry of the layers in our new Vp seismic velocity model is guided by MCS imaging of sediments, magmatic sills, and basement along SCREECH Line 2 [Shillington et al., 2006; Péron-Pinvidic et al., 2010] (Figure 3d). Shillington et al.  identified two distinct sediment layers above the crystalline basement in their MCS image that are divided by horizon AU. At larger depth in the COT, Péron-Pinvidic et al.  distinguished an interval of anomalously bright reflections topped by the so-called U reflection. This ~1200 m thick sequence lies between the flat basement and overlying sediments, and is interpreted as a package of postrift sills. This interpretation differs from that of Van Avendonk et al. [2006, 2009] whose P-velocity model for SCREECH Line 2 assumed a negligible thickness for the postrift sills in the COT. Basement beneath postrift sills between 180 and 240 km model distance is relatively flat when compared to the highly variable topography of faulted basement at the seaward end of SCREECH Line 2.
 In the OBS records of SCREECH Line 2 we interpret three pairs of reflected and refracted phases that appear to be consistent with the layering we describe in the MCS image. We assign four compressional refraction (P1–P4) and three reflection (R1–R3) arrivals in the wide-angle data. As the earliest-arriving phase present in all OBS records, P4 increases in apparent velocity from 5.0–6.0 km/s at 15–20 km offset to 6.0–8.5 km/s at offsets up to 150 km (Figures 2, A2–A5). Local variations in P4 apparent velocity mimic MCS basement relief. Refraction P3 is a straight travel-time branch in x-t space and is typically observed at source-receiver offsets of 10–15 km. Apparent velocities for P3 are 2.2–3.2 km/s, with most being slightly higher than 3.0 km/s, on par with the average velocity reported for postrift sill and sediments [Péron-Pinvidic et al., 2010]. The slowest refracted arrivals P2 and P1 intersect the direct water wave with apparent velocities of 1.9–2.5 km/s and 1.7–2.2 km/s, respectively, and their amplitudes diminish at larger offsets. The x-t diagrams also show that at short source-receiver offsets the seismic refraction P4 is asymptotic to the wide-angle seismic reflection R3, and both arrivals are observed from each instrument. Limited to instruments above the postrift sills, refraction P3 is asymptotic to reflection R2. Refraction P2 is asymptotic to reflection R1.
 We picked 6647 arrival times of P-waves from air-gun shots recorded on 17 OBSs between the continental slope and the seaward end of SCREECH Line 2, and 1213 reflection travel times in the coincident MCS data. We assigned uncertainties to these travel times based on pick and data quality. After carrying out the inversion of these data using equation (1), we obtained a new P-velocity model that has a low root-mean-square misfit (RMS = 116 ms) and a χ2 value of 1.0.
4 Shear-Wave and Vp / Vs Modeling
 We picked 863 arrival times from shear waves (S) turning beneath the basement surface to obtain different constraints on basement structure and composition. Not all of the SCREECH Line 2 instruments have observable S-waves (Figure 3a), so the ray coverage in our S-wave inversion is more sparse compared to the P-wave inversion. S-waves have apparent velocities ranging from 3.2 to 4.5 km/s, much lower than phase P4, and often have a variable slope in x-t space due to basement depth variations, distinguishing them from compressional refractions with similar (P3) and slightly slower (P1 and P2) velocities.
 A challenge with using shear-wave travel times is the fact that the depth of conversion from P- to S-wave energy in the downgoing raypath is not known. Such conversions may happen at the seafloor or on a deeper boundary with significant impedance contrasts [Mjelde et al., 2003; Eccles et al., 2009]. Based on inspection of the S-wave arrivals from SCREECH Line 2, we find that the majority of these observations must represent phases that converted at the top of the basement. We use this assumption to calculate travel-time residuals for the S-wave picks and invert the data for a Vp / Vs ratio in the basement, applying equation (2) where we use the Vp model from the tomographic inversion of the previous section. After a few iterations of the raytracing and linearized inversion steps, we obtain a Vp / Vs ratio model (Figure 3c) with a low RMS misfit of 135 ms. It is difficult to quantify the uncertainty in our Vp / Vs model from a statistical standpoint given the paucity of shear-wave data. However, we understand our models to be reasonably well-constrained with average Vp / Vs uncertainties between 0.1 and 0.2 based on comparisons of bulk Vp / Vs models (Figure A1).
 To better estimate the lithology of the Newfoundland COT, we plot four depth profiles of Vp against Vp / Vs (Figure 4). Key model distances of 185, 210, 250, and 315 km (Figure 3c) are plotted in an attempt to characterize the most outboard remnant of probable upper continental crust, the region beneath the flat COT basement, the onset of basement with variable topography near ODP 1277, and the seaward end of SCREECH Line 2, respectively. Velocities are plotted every 0.2 km depth for all model space that is constrained by shear waves. SCREECH Line 2 velocities are plotted above established ranges of Vp and Vp / Vs for common COT lithologies (Table A1). For the middle and lower continental crust we plot the felsic rocks quartzite and granite and the mafic rocks gabbro and granulite [Holbrook et al., 1992]. We also plot seismic velocities of basalts found in older oceanic crust [Hyndman, 1979; Brocher, 2005] and mantle peridotites with 0–100% serpentinization [Christensen, 2004].
 Between 130 and 180 km model distance on SCREECH Line 2 (Figure 3), low Vp / Vs ratios (1.57–1.65) are consistent with previous interpretations that this region represents the last portion of the upper continental crust on the Newfoundland margin, including a small package of prerift sediments at ~170 km model distance [Shillington et al., 2006; Van Avendonk et al., 2006]. The uppermost 3.5 km of basement at 185 km model distance (Figure 4, blue diamonds) plots Vp and Vp / Vs consistent with quartzite and granite [Holbrook et al., 1992], supporting the interpretation that continental crust with a predominantly felsic composition typifies this inboard section of the Newfoundland margin.
 The most enigmatic aspect of our seismic velocity models is the uppermost 3–5 km of flat COT basement between 180 and 240 km model distance where seismic velocities are relatively homogeneous (Vp = 5.0–6.2 km/s; Vs = 3.0–3.3 km/s) compared to those in the shallow, rough basement immediately seaward. The Vp in the flat COT basement is low enough to be consistent with continental crust [Holbrook et al., 1992], as is suggested by Van Avendonk et al. . Other workers contend that the COT in the Newfoundland Basin comprises slow-spreading oceanic crust [Srivastava et al., 2000] or exhumed mantle [Reid, 1994; Sibuet et al., 2007]. Our recovered Vp / Vs ratios alone cannot rule out the possibility proposed by Srivastava et al.  that the Newfoundland basement is capped by thin ocean crust, although we note that Vp in the flat COT is lower than what we expect for ocean crust (5.0–7.5 km/s). In this case, weak magnetization in the COT used to support the presence of oceanic crust (e.g., M0 and older; Figure 3d) may be masked by postrift sills above the flat basement [Péron-Pinvidic et al., 2010]. Although zones of exhumed continental mantle often have Vp around 7.6 km/s [Dean et al., 2000; Whitmarsh et al., 2001], mantle peridotites range in Vp from 4.8 to 8.0 km/s depending on the extent of serpentinization [Christensen, 2004]. Thus, serpentinized peridotite is a viable explanation for the 5.5 km/s shallow COT basement in the Newfoundland Basin if exhumed mantle rock here is 75% serpentinized. Such a large degree of serpentinization, however, would require S-wave velocities as low as 2.75 km/s (Vp / Vs = 2.0). We find that Vp / Vs ratios between 1.67 and 1.77 in the uppermost basement of the COT are too low to primarily represent serpentinized peridotite (Figure 3c). Plots of Vp versus Vp / Vs at 210 km model distance show the top ~2 km of the flat-lying basement approaching velocities of felsic continental crustal rocks (Figure 4, red squares). As Vp increases to >6.5 km/s at depths >5 km beneath the top of basement, the 210 km profile plots more closely to velocities of mafic rocks such as gabbro and/or 10–25% serpentinized mantle peridotite. These results are consistent with interpretations of the uppermost COT basement as unroofed and stretched continental lower crust overlying hydrated mantle peridotites [Van Avendonk et al., 2006].
 The shallow, rough basement at 250 km model distance between ODP Sites 1276 and 1277 exhibits seismic properties (Vp = 4.0–6.0 km/s; Vp / Vs = 1.72–1.83) that are inconsistent with serpentinized mantle as an average composition (Figure 4, green triangles), although such ultramafic rocks (along with minor amounts of basalt and gabbro) were drilled at Site 1277 [Robertson, 2007]. Our model instead plots velocities in the uppermost ~3 km of basement between oceanic basalt and granite, which could indicate the presence of continental rocks far seaward in the COT and/or increasing amounts of volcanics in the shallow basement emplaced during the gradual transition from rifting to seafloor-spreading. Poisson's ratios indicating shallow granodioritic crust (Vp / Vs = 1.78) were similarly found northeast of the Flemish Cap, and a seaward increase in Poisson's ratios (Vp / Vs = 1.86–1.89) near a ridge-like structure were interpreted as a combination of serpentinized mantle and volcanic material [Gerlings et al., 2011]. At depths >5 km below the top of SCREECH Line 2 basement, the 250 km depth profile exhibits velocities (Vp >7.7 km/s; Vp / Vs = 1.78–1.88) that match the seismic properties of hydrated mantle rocks and may indicate a decrease in peridotite serpentinization from 25% to 0% with increasing depth.
 The most seaward section of SCREECH Line 2 is relatively poorly constrained by the S-wave data (Figure 3), and thus the dominant presence of either oceanic crust or serpentinized mantle is equivocal. Furthermore, we cannot rule out gabbroic underplating and melt infiltration at the seaward end SCREECH Line 2 [Bronner et al., 2011], which would perhaps be consistent with a gradation from exhumed lithosphere to “incipient” oceanic crust of intermediate composition. The seaward-most Vp versus Vp / Vs plot at 315 km exhibits seismic velocities between granite and oceanic basalt at shallow basement depths, and approaches velocities of minorly serpentinized (0–10%) mantle peridotites at depths of 5 km below basement (Figure 4, purple circles). We caution that plots of seismic properties here may be artifacts of insufficient shear-wave raypath coverage. To image the onset of “true” oceanic crust with normal 6–7 km thickness, additional seismic reflection and refraction data need to be acquired seaward of SCREECH Line 2.
 Our analysis of Vp / Vs ratios from SCREECH Line 2 supports the hypothesis that highly stretched continental lower crust extends far into the Newfoundland Basin COT. Development of flat basement in the COT was likely shaped by low-angle, landward-dipping detachment faults, although postrift volcanic sills mask the observation of such large-scale faults in MCS data. Shear velocities support the presence of exhumed mantle deep beneath the flat COT basement and around ODP Site 1277, where extensive basement faulting led to partial serpentinization of peridotites that subsequently were infiltrated and overprinted by melt products from the upwelling asthenosphere. Mantle hydration is suggested to decrease with depth. Continued plate divergence between Newfoundland and Iberia led to increasing volcanism that eventually emplaced normal thickness oceanic crust in the Atlantic, but this transition is either not observed at SCREECH Line 2 or is unable to be distinguished from available seismic velocity models.
 To investigate how well S-wave arrivals can constrain Vp / Vs in the COT (and thus distinguish between different possible end-member lithologies), we test how models with three specific Vp / Vs ratios affect S-wave travel-time misfits for OBS 15, 17, and 19 in the Newfoundland COT (Figure A1; (a) Vp / Vs = 1.65; (b) Vp / Vs = 1.77; (c) Vp / Vs = 1.87). Blue and red lines represent picked and calculated S-wave travel-time curves, respectively. Reduction velocity is 4.0 km/s. To test for the possible presence of felsic continental upper crust, we use a bulk Vp / Vs ratio of 1.65. For more mafic-rich rocks found in oceanic crust or in lower continental crust (either naturally or as exhumed underplated material), we assume Vp / Vs = 1.77 [Brocher, 2005]. Alternatively, for mantle peridotites that are 25% serpentinized we assume Vp / Vs = 1.87 [Christensen, 2004]. In a few raytracing tests, we keep Vp fixed and locally vary the Vp / Vs ratio to determine how it affects the data fit for shear waves that best sample the flat COT basement. The bulk model with Vp / Vs = 1.77 gives the best fit of S-wave travel times (RMS = 124 ms). The model with Vp / Vs = 1.87 has a higher misfit (RMS = 183 ms), and a Vp / Vs ratio of 1.65 fits very poorly (RMS = 322 ms). Although not shown in Figure A1, additional testing of bulk Vp / Vs models found that landward of the COT, a low Vp / Vs ratio (1.65) fits the S-wave data well, while higher ratios fit poorly. This supports the hypothesis that felsic continental upper crustal rocks characterize the Newfoundland margin here. At the seaward end of SCREECH Line 2, bulk models with Vp / Vs = 1.87 and Vp / Vs = 1.77 produce the lowest misfits (110 and 104 ms, respectively), and the nature of the lithosphere is equivocal given our data.
 Appendix figures A2–A5 show receiver gathers for OBS 08, 12, 19, and 23. Refraction and reflection arrivals are described in the text.
Table A1. Compressional and Shear Wave Velocities of Common COT Lithologiesa
Error bars for COT rock types are displayed in Figure 4.
 D.R.E. was supported by the Maurice Ewing and J. Lamar Worzel Graduate Student Fellowship from The University of Texas Institute for Geophysics (UTIG contribution #2572). The authors thank Keith Louden and an anonymous reviewer for their comments.
 The Editor thanks Keith Louden and an anonymous reviewer for their assistance in evaluating this paper.