A program of explosion seismology in central North Island, New Zealand, discovered a strong reflector within the upper mantle. Reflections from this (PmP2) are spatially confined to come from an interface 35 km deep and directly beneath a 40 km-wide, back-arc extension zone with active volcanism, high heat flow, low Pn wave-speeds and thinned crust. On the basis of relative reflection amplitudes, the mantle reflections are most readily explained by an interface with a negative seismic impedance contrast. A satisfactory fit is obtained for a layer with a 40–90% drop in S-wave speed (Vs) compared to the surrounding mantle. We interpret this layer to be a 40 km-wide reservoir of partial melt pooled at a thermal boundary layer within the upper mantle.
 High amplitude, or “bright spot”, reflections in the crust have been discovered within many volcanic and extensional terranes [Sheetz and Schlue, 1992]. Marine seismic methods have been particularly successful in identifying bright spot reflections as coming from melt bodies in the crust and mantle of oceanic spreading centers [Singh et al., 1998], but bright spot reflections from within the continental mantle are less common [Marsh, 1989]. This study documents an unusually strong mantle reflector beneath central North Island, New Zealand, and interprets it as coming from the top of a layer of partial melt. Such an interpretation is in keeping with the back-arc setting of the central North Island above an active subduction zone [Darby et al., 2000]. This is important because although it is now widely accepted that melts are generated in the mantle wedge above subducted slabs, it is still unclear how melt is transported through the wedge and where it pools and differentiates [Davies and Bickle, 1991].
 Pliocene back-arc spreading from the Havre Trough to the north (Figure 1) has advanced into the continental lithosphere of New Zealand [Karig, 1970], resulting in a wide range of volcanic and geothermal phenomena [Cole et al., 1995]. Approximately 12,000 km3 of silicic, Quaternary volcanics are present within the Central Volcanic Region (CVR) of the North Island (Figure 1) in a 2 km-deep graben [Stern and Davey, 1987]. Geothermal fields discharge heat at the rate of ∼4.3 × 109 W from the eastern side of the CVR [Bibby et al., 1995], which corresponds to a heat output, per unit strike length of volcanic zone, of 27 MW/km; a value similar to an oceanic spreading setting like Iceland [Bodvarsson, 1982].
3. Crustal Structure
 Nine land-based explosions on two perpendicular profiles (Figure 1) were shot as part of a crust and upper-mantle velocity study of the central North Island [Henrys et al., 2003]. The two roughly perpendicular seismic profiles cross the wedge-shaped CVR and the flanking terranes of Mesozoic greywacke, which constitutes basement rock for New Zealand. Ignimbrites and other volcanics extend to a depth of ∼2.0 km in the upper crust and have seismic velocities of 1.5–3.0 km/s (Figure 2b). Below these superficial volcanics, velocities (Pg phase) are ≤6 km/s. These velocities are, in general, 6% lower than those each side of the CVR at a comparable depth (Figure 2b; see auxiliary material).
 The fastest velocity recorded along line 1 is 7.35 ± 0.08 km/s and has a cross-over distance [Sheriff and Geldart, 1995], with respect to Pg, of about 85 km (Figure 2c). We interpret these as reduced upper-mantle (Pn) velocities, rather than lower crustal for the following reason. A Pn of 7.4 ± 0.1 km/s is reported for the central North Island, on the basis of earthquake travel-times between station pairs [Haines, 1979]. Offsets for New Zealand crustal earthquakes from the seismograph stations used in this analysis were up to 750 km, which requires the 7.4 km/s layer to extend to a depth of at least 80 km (see auxiliary material).
 Other continental rifts are also observed to have Pn speeds reduced by 10% and to depths of 100 km or more [Davis and Slack, 2002]. A Pn wave-speed of 7.4 km/s is about 10% less than normal [Christensen and Mooney, 1995]. Relating a reduction in Pn to temperature and degree of partial melt is complex as the rate of decrease is strongly dependent on the shape of melt inclusions [Hammond and Humphreys, 2000]. Nevertheless, based on laboratory studies of realistic shaped melt inclusions, a 10% drop in Pn can be ascribed to about 2% of partial melt [Hammond and Humphreys, 2000].
 A first-order estimate of crustal thickness in terms of a single plane-layer can be obtained from the cross-over distance [Sheriff and Geldart, 1995] between the Pg and Pn segments of a travel time plot (Figure 2c). For an average crustal velocity of 5.7 km/s, 7.4 km/s for the upper mantle and a cross-over distance of 85 km (Figure 2c) a crustal thickness of 16 ± 1 km is estimated. This same ∼16 km deep interface seen in the refraction data of line 1 is also identified by reflections along line 2 from shots each side of the CVR (Figure 2a; see auxiliary material). These reflections, here termed PmP1, define an arched interface between a depth of 16 to 20 km and mark the change from less than 6.0 km/s to greater than 7.0 km/s velocities as determined from line 1 (Figure 2c; see auxiliary material).
 A thinned crust beneath the CVR concurs with the results of previous seismic exploration in central North Island [Stern and Davey, 1987], and with the present-day extension rates of 8 ± 2 mm/y [Darby et al., 2000]. Accordingly, the start of a crust-mantle transition zone is interpreted to be at a depth of 16 ± 1 km. This is one of the thinnest continental crusts reported but is similar to that of the Afar region of Ethiopia [Christensen and Mooney, 1995] and some extended continental margins [Ansorge et al., 1992]. Given that the average crustal thickness of the New Zealand continent is less than normal this result is not implausible. For example, “stable” regions of New Zealand well to the north or east of the present plate boundary have crustal thicknesses measured by seismic methods of between 25 and 27 km [Stern et al., 1987; Reyners and Cowan, 1993]. Thus, assuming uniform extension, a 16 km thick crust implies an extensional factor of β = 1.5–2.
4. Mantle Reflector
 Reflections from a second reflector (PmP2) arrive at offsets between 80–120 km from an interface at a depth of 35 ± 3 km beneath the CVR (Figures 2b and 2d). We term this the Taupo mantle reflector and note that it is constrained by shots from both east and west on line 2, to be within 5° of horizontal (Figure 2b; see auxiliary material). This reflector is remarkable for three reasons: it produces high amplitudes compared to Pg and PmP1; it is below the 16 km deep crust-mantle interface and is therefore within the upper mantle; and it is confined in lateral extent (∼40 km) to be beneath the active volcanic area where crustal extension rates are highest [Darby et al., 2000] (see auxiliary material).
 Reflection amplitudes are dependent on impedance contrast and incidence angle, but also attenuation (Q) along the travel path, thickness of the layer and local source and receiver conditions [Sheriff and Geldart, 1995]. A useful method to allow for source and receiver conditions, and attenuation is to normalize the amplitude of PmP against Pg on a single trace [Auger et al., 2003]. On this basis, the ratio PmP2/PmP1 = 1.5 ± 0.5 is estimated (Figure 3a; see auxiliary material). Taking into account the travel path through PmP1, for rays to and from PmP2, this estimate is a minimum (see auxiliary material).
 PmP2 displays strongest amplitudes for incidence angles between 50 and 65° with a peak at about 63° (Figure 3b). For this angle range it is difficult to explain PmP2 as a Moho reflection where P-wave velocity increases from say 7.4 to 8.1 km/s. Based on Zoeppritz's equations for uniform plane layers [Sheriff and Geldart, 1995], negligible reflection energy is predicted for this incidence angle range, but maximum energy for angles >66° (Figure 3b). We observe the opposite variation in amplitudes with incidence angle (Figure 3b); a broad increase of amplitude from an incidence angle of 50° to a peak around 63° then a decline in amplitude. This variation is best explained by a family of solutions that have a negative impedance contrast at PmP2, mainly due to a large drop in Vs. In particular, to get both the strongest amplitude response, and a peak between 60–63° we find the best solution to be a 90% drop in Vs and a 5% drop in Vp (Figure 3b). A 40% drop in Vs will also give peak amplitudes in the required incidence-angle range, but the predicted amplitude is only half of that for a reflection from a body with a 90% drop in Vs. Although non-unique, a 40–90% drop in Vs gives a simple way of explaining the amplitude characteristics of PmP2. Such a drop in Vs is indicative of the presence of fluids or melt, and for realistic melt geometries will predict about a 6–12% partial melt [Hammond and Humphreys, 2000].
 Tuning may enhance reflection amplitudes if the PmP2-reflector is thin, with respect to the dominant seismic wavelength. However, the present data set is not sufficient to resolve layer thickness. But if we are detecting the top of a melt or fluid layer, its lower boundary is likely to be gradational and may not be seen seismically [Marsh, 1989].
 Similar results to those reported here have been reported from beneath other volcanic provinces such as the Rio Grande rift [Sheetz and Schlue, 1992], Mt Vesuvius [Auger et al., 2003] and the central Andes [Chmielowski et al., 1999]. These studies use P to S conversions to propose bodies with Vs < 1.0 km/s. What makes the proposed melt body beneath the CVR different from the above examples is that it is in the mantle. Melts in the mantle are expected given the widely held view that magmas in back-arc systems are derived from the mantle wedge; water from dehydration of the slab lowers the mantle solidus [Davies and Bickle, 1991; Stern, 2002]. That the Taupo mantle reflection is only seen directly below the region of crustal thinning (Figure 2a) suggests, that decompression melting could also be exerting a critical control on the lateral and vertical extent of magma pooling beneath the CVR.
 One interpretation of the Taupo reflection is that it comes from the top of a rising diapir of mantle melt. Such phenomena are predicted to have ascent rates of perhaps ∼1 m/y [Stern, 2002] up to the crust-mantle boundary where they differentiate into mafic and silicic components. Alternatively, the boundary at a depth of 35 km maybe static, and represents a thermal boundary layer (i.e., the base of the lithosphere) where a viscosity contrast inhibits the rise of melts (Figure 4). But regardless of whether the Taupo mantle reflector represents a static or dynamic phenomenon, it is a likely candidate for controlling melt and hence volcanism seen at the surface.
 This data set was collected as part of the NIGHT (North Island Geophysical Transect) project. This was a collaborative effort between institutions in New Zealand, UK and Japan [Henrys et al., 2003]. We thank M. Savage, D. Walcott, K. Furlong, E. Smith and A. Galvé for constructive comments on an initial version of this paper.