The Agulhas Leakage represents a significant portion of the warm, surface return flow of the global overturning circulation and thus may be an important feedback in the ocean climate system. Models indicate that reduced leakage could be caused by a stronger Agulhas Current and/or a more upstream (eastward) Agulhas Retroflection, while a weaker Agulhas Current would result in a more westward retroflection and increased leakage. However, data for the Last Glacial Maximum support both a weaker Agulhas Current and less leakage, implying a possible displacement of the retroflection. We present new 87Sr/86Sr results for modern sediments within this region, confirming that the modern pathway of the Agulhas Current, Retroflection, and Leakage can be traced by terrigenous sediment provenance using Sr isotopes. New 87Sr/86Sr data from sediments deposited during the Last Glacial Maximum suggest that the glacial Agulhas Current and Retroflection followed nearly their modern trajectory. The provenance data appear to rule out both a stronger Agulhas Current and a more upstream Agulhas Retroflection. We conclude that the reduced glacial leakage was caused by the weakened Agulhas Current, with no significant change in the retroflection position. This is inconsistent with the model predictions and thus emphasizes the need for further work in this region.
 It has also been postulated that the latitude of the Agulhas Retroflection is partly controlled by the position of the Subtropical Front (STF). Models suggest that a southward displacement of the zonal wind profile in a one-layer model tends to increase the Agulhas Leakage [de Ruijter and Boudra, 1985], while a 5° northward displacement of the zero of the wind stress curl tends to isolate the south Indian circulation from the Atlantic gyre [Matano, 1996]. Glacial reconstructions of the STF south of Africa range from little to no change [Hays et al., 1976] to northward migrations of a few degrees [Howard and Prell, 1992; Peeters et al., 2004; Prell et al., 1980; Trend-Staid and Prell, 2002; Verardo, 1995]. Until now, there have been no data between 36.5 and 40.5°S to constrain the mean position of the Agulhas Retroflection during the LGM or during the late Pleistocene in general. The goal of this study was to constrain the position of the Agulhas Retroflection at the LGM using strontium isotope ratios (87Sr/86Sr) of the terrigenous fraction of deep-sea sediments.
 Many prior studies have demonstrated the validity of using 87Sr/86Sr of terrigenous detritus to infer their provenance and relate them to changes in ocean circulation patterns [Biscaye and Dasch, 1971; Eisenhauer et al., 1999; Franzese et al., 2006; Hemming et al., 2007; Revel et al., 1996a, 1996b; Walter et al., 2000]. Our own previous work has shown that sediments transported by the Agulhas Current have high 87Sr/86Sr derived from the old continental terrains on the eastern coast of Africa [Franzese et al., 2006]. We used this information, along with other provenance and flux data, to determine that the Agulhas Current's sediment transport capacity was significantly reduced (i.e., the Agulhas Current was weaker) during the LGM [Franzese et al., 2006]. Additionally, the pattern of terrigenous 87Sr/86Sr of Holocene sediments deposited in the region south of Africa strongly resembles the surface ocean circulation such that sediments underlying the Agulhas Current and the Agulhas Return Current have higher 87Sr/86Sr than the surrounding sediments, and the high 87Sr/86Sr clearly traces the modern path of the Agulhas Current as well as its retroflection and leakage into the Atlantic [Franzese et al., 2006]. We therefore infer that the region of high 87Sr/86Sr can be used to trace the mean path of the Agulhas Current, Retroflection, and Return Current in the past.
2. Core Samples
 The chosen core sites are noncontourite sites from four distinct zones within the greater Agulhas Retroflection region (Table 1 and Figure 2a). These are directly beneath the Agulhas Current from 29 to 35°S and water depths between 1 and 4 km, the deep Transkei Basin (∼4.5 km, near 35°S, 29°E), the retroflection region between about 36 and 40°S and 19 and 26°E and water depths between 1 and 5 km, and a 3–4 km depth transect off the Agulhas Plateau. Two deeper (5–6 km) “Southern” cores were also sampled near 45°S. The deep waters in this area are North Atlantic Deep Water (NADW) and Circumpolar Deep Water (CPDW) centered around 3 km and Antarctic Bottom Water (AABW) below about 4 km (Figure 2b). We then sought fast, straightforward means of identifying (1) the depth of the LGM horizon and (2) the main provenance of each core.
Table 1. New Strontium Isotope Data for the Holocene and LGMa
The abbreviations for stratigraphic tools are as follows: 14C, radiocarbon analyses performed by NOSAMS; 14C(ext), extrapolation between 14C tie points; MS, magnetic susceptibility (this study); CaCO3, the percentage of CaCO3; MP, micropaleontology (shipboard analysis of foraminiferal assemblages [Hall and Zahn, 2004]); Color, color changes correlated to those in CD154-06-6PK; Lithics, counts of lithic grains >63 μm (this study); δ18O, the δ18O measured on benthic forams (provided by U. Ninnemann); C. davisiana, stratigraphy based on radiolarian abundances [Hays et al., 1976].
Record was provided by L. Burckle (unpublished data, 2004).
 Core lithologies are highly varied, ranging from foraminiferal ooze to silty lutite. Accordingly, we employed several different methods for determining core stratigraphies. Our preliminary stratigraphic determinations for cores from the Lamont-Doherty Earth Observatory (LDEO) Deep-Sea Sample Repository used core descriptions from the cruise logs, percent calcium carbonate, counts of lithic grains per gram, and magnetic susceptibility (Table 1 and auxiliary material Data Sets S1–S12 and Tables S1–S14). We also include cores collected during RRS Charles Darwin Cruise 154 (CD154), for which some preliminary stratigraphic determinations were made on board the ship [Hall and Zahn, 2004]. Where available, we made use of published stratigraphic information [Hays et al., 1976; Verardo, 1995] as well as unpublished % CaCO3 data courtesy of L. Burckle (2004) and unpublished δ18O data courtesy of U. Ninnemann (2002). When possible, the depth of the LGM horizon (defined as 19–26 ka) was confirmed with 14C dating. The resolution, accuracy, and precision of the age models are highly variable, simply because there is a great deal of variability in data available from core to core. A summary is provided in Table 1, and details for the stratigraphic determinations of each core are in the auxiliary material and will be made available online through the World Data Center (WDC) for Paleoclimatology (http://www.ncdc.noaa.gov/paleo/paleocean.html).
3. Analytical Procedures
 All Sr isotope measurements are on the <63 μm terrigenous sediment fraction. Calcium carbonate and ferromanganese oxides were sequentially removed using a buffered acetic acid leach (pH = 5) followed by 0.02 M hydroxylamine HCl in 25% acetic acid [Rutberg et al., 2005].
 In the remaining terrigenous fractions, ∼10 mg were dissolved for Sr isotope analyses. The presence of organic material and/or precipitated fluorides often caused difficulty when dissolving samples. Samples were dissolved on a hot plate in 4 mL of 8 N nitric acid (HNO3) and 200 μL of hydrofluoric acid (HF), which was repeated once or twice after sonicating for 20–30 min prior to heating. Aqua regia heated to ∼170°C was also used up to three times when necessary. Each sample was dried down in a small amount of concentrated HNO3 before redissolving in 3 N HNO3 for Sr separation using established LDEO procedures [Rutberg, 2000; Rutberg et al., 2005].
 Isolated Sr was loaded on tungsten filaments with a tantalum chloride solution [Birck, 1986]. Sr isotope ratios were measured at LDEO by dynamic multicollection using a VG Sector 54–30 thermal ionization mass spectrometer (TIMS) and corrected for mass fractionation by normalizing to 86Sr/88Sr = 0.1194. Typical beam intensities were 2.0–6.0 × 1011 A on mass 88. Each analysis reflects ∼80 ratios. External reproducibility was monitored using the NBS-987 Sr standard, which gave 87Sr/86Sr = 0.710270 ± 20 (2σ reproducibility, n = 141). All data are corrected to 87Sr/86Sr = 0.71025 for NBS-987.
 Forty-one new data add to the existing set of 13 Holocene terrigenous 87Sr/86Sr between 28–44°S and 17–34°E [Franzese et al., 2006]. All the data show high 87Sr/86Sr (0.7203–0.7360) under the path of the Agulhas Current, in the retroflection region, and on the Agulhas Plateau (which lies under the path of the modern Agulhas Return Current) and much lower values in the Transkei Basin (87Sr/86Sr = 0.7137) and south of 43°S (87Sr/86Sr = 0.7151 and 0.7092) (Table 1 and Figure 3a). The data exhibit a general trend of decreasing 87Sr/86Sr with increasing water depth, and beneath the Agulhas Current, 87Sr/86Sr generally decreases downstream (i.e., from NE to SW).
 The data presented here for the LGM add 19 new data points to the preexisting set of only 5 in this region. The range of 87Sr/86Sr in LGM samples underlying the modern path of the Agulhas Current is 0.7159–0.7308 (Table 1 and Figure 3b). In the retroflection region, LGM values are slightly lower (0.7161–0.7192). Glacial sediments with 87Sr/86Sr ≥ 0.720 reach as far south as 36.5°S off the east and west coasts of South Africa, while south of 43°S values are much lower (87Sr/86Sr = 0.7139 and 0.7123). In the retroflection and Agulhas Plateau regions, LGM 87Sr/86Sr are consistently lower than Holocene values.
 Excluding two cores with glacial 87Sr/86Sr ∼ 0.716, the lowest 87Sr/86Sr north of 43°S are found in the Transkei Basin. The glacial samples from the Transkei Basin have 87Sr/86Sr ∼ 0.717, which is higher than the one Holocene sample from that region (87Sr/86Sr = 0.7137) but still lower than the glacial 87Sr/86Sr of most cores beneath the present-day Agulhas Current and Retroflection (Table 1 and Figure 3b). Additionally, the Holocene value of 0.7137 is exceptionally low compared to other cores in this region (Figure 3a).
 Five cores from the flanks of the Agulhas Plateau have extremely consistent terrigenous 87Sr/86Sr for both the Holocene and LGM. These cores make up a NW to SE transect spanning water depths from ∼3 to 4 km (Table 1 and Figure 4). Three (VM34-153 and VM34-158 (this study) and VM34-157 [Franzese et al., 2006]) are well within the modern CPDW, one (VM34-155) is well within modern AABW, and one (VM34-156) is located very close to the interface (∼3.8 km [Arhan et al., 2003]). Their Holocene terrigenous 87Sr/86Sr ranges from 0.7220 to 0.7251, while LGM values fill an even smaller range, from 0.7177 to 0.7189. There is no trend in terrigenous 87Sr/86Sr with water depth for either time slice (Figure 4).
5.1. Retroflection Location
 The main subject we address here is whether a change in the mean location of the Agulhas Retroflection was responsible for the reduced Agulhas Leakage during the LGM. Alternatively, if the Agulhas Retroflection occurred in the same place as it does today, the reduced glacial leakage must have been caused by a weaker flow through the Agulhas Current system.
 Terrigenous 87Sr/86Sr of Holocene sediments clearly show that in the region south of Africa, 87Sr/86Sr ≥ 0.720 trace the mean integrated flow path of the modern Agulhas Current through the retroflection and into the Agulhas Return Current (i.e., the Agulhas Plateau) (Figure 3a). We therefore infer that we can use terrigenous 87Sr/86Sr ≥ 0.720 to identify the retroflection's average location in the past. The interpretation of changes in terrigenous 87Sr/86Sr as provenance variability can potentially be complicated by the processes of weathering and transport. For sediments in the Agulhas Current and Retroflection region, these influences were inferred to be minor on the basis of apparently unchanging end-member compositions over the relevant timescale, the consistency of change over a broad spatial scale spanning a large range in sedimentary and oceanographic settings, and comparisons with Rb-Sr and the Sm-Nd isotope system [Franzese et al., 2006]. Major and trace element concentrations, sortable silt mean size, and 87Sr/86Sr and trace element data for grain size separates even more firmly show that the primary control on the terrigenous 87Sr/86Sr of deep sea sediments in this region is provenance, with only minor secondary modification by sedimentary processes such as sorting [Franzese, 2008]. We are therefore confident that 87Sr/86Sr ≥ 0.720 can be used to trace the path of the Agulhas Retroflection location through time. In the following paragraphs, we test and rule out several hypotheses for the position of the glacial Agulhas Retroflection, leaving only one hypothesis uncontradicted: no glacial to Holocene change in the location of the Agulhas Retroflection.
 The first case we consider is an extreme migration of the Agulhas Retroflection to a position northeast of the Agulhas Plateau (near 35°S, 28°E), with the STF reaching far north of its present range, almost touching the tip of Africa (Figure 5a). Floral assemblages and planktonic δ18O from a suite of cores underlying the Agulhas Current indicate that the main flow path of the Agulhas Current was similar to today at least as far south as 34°S and probably as far as 36°S [Winter and Martin, 1990]. Basin-wide sea surface temperature reconstructions based on faunal assemblages are mostly consistent with a northward shift of the STF of only a few degrees at the LGM [Hays et al., 1976; Howard and Prell, 1992; Prell et al., 1980; Trend-Staid and Prell, 2002]. The strontium isotope results we present here provide additional constraints for the LGM Agulhas Current system. In particular, if the retroflection was near 35°S, 28°E during the LGM, we should expect to see an increased Agulhas signature in the terrigenous provenance (i.e., higher than modern 87Sr/86Sr) in that region, along with a decreased Agulhas signature (i.e., lower than modern 87Sr/86Sr) beneath the modern retroflection near 40°S, 20°E.
 Although our coverage is sparse in the key region east of the Agulhas Plateau, we do not see exceptionally high 87Sr/86Sr in the Transkei Basin during the LGM. Holocene and late Pleistocene sediments are scarce in this area because of scouring by bottom currents, and the deepest parts of the basin are well below the lysocline, so there is too little carbonate material to establish a reliable stratigraphy. Consequently, we have only two 87Sr/86Sr data points for the Transkei Basin. On the basis of the relatively low glacial 87Sr/86Sr of these two cores (∼0.717) compared to that of cores beneath the modern path of the Agulhas Current and Retroflection (0.7159–0.7308), however, we infer that the Agulhas Retroflection was not positioned east of the Agulhas Plateau during the LGM.
 In the modern situation, with the Agulhas Retroflection located west of the Agulhas Plateau, Agulhas-derived particles can simply be carried south of the Agulhas Plateau by the surface waters of the Agulhas Retroflection and the Agulhas Return Current. In addition, Agulhas-derived particles rained out in the vicinity of the retroflection can be carried south of the Agulhas Plateau by deep and bottom currents (Figure 2). If the retroflection was as far north as 37°S during the LGM, only the counterclockwise flowing AABW could transport Agulhas-derived particles to the southern Agulhas Plateau (Figure 5b). The slightly shallower NADW and CPDW tend to flow eastward, so they could not be responsible for bringing Agulhas-derived material to these core sites unless the retroflection was near its modern position. Our glacial results show that the cores from CPDW depths have the same provenance as those affected by AABW (Figure 4). The vertical homogeneity of LGM 87Sr/86Sr most likely reflects a surface water source that is common to all sites. This could only be true if, during the LGM, the Agulhas Retroflection was positioned west of the Agulhas Plateau, close to its modern position.
 Therefore, on the basis of the available data, the most plausible scenario for the LGM circulation is no change from the modern flow path of the Agulhas Current or the location of retroflection (Figure 5c). The glacial pattern of terrigenous 87Sr/86Sr in the sediments south of Africa is very similar to today, with a region of higher 87Sr/86Sr beneath the modern path of the Agulhas Current and Retroflection. The major difference we observe between the Holocene and LGM sediments is lower glacial 87Sr/86Sr for the Agulhas region as a whole (Figure 3), implying a smaller relative input of old continental material during the LGM but no major difference in the circulation patterns.
5.2. Agulhas Current Strength
 Combining our new 87Sr/86Sr data with what we know from previous studies, however, we find compelling evidence for a reduced input from the Agulhas Current during the LGM. The data imply that during the LGM, the Agulhas Current flowed along its modern trajectory but deposited less material with very high 87Sr/86Sr. In the retroflection region, the Agulhas Current–derived, high-87Sr/86Sr material mixes with the lower-87Sr/86Sr ambient sediment to give each sample its ultimate terrigenous 87Sr/86Sr. The ambient sediment includes minor contributions from the local south African continent but is mainly derived from the southwest Atlantic and delivered by deep and bottom waters [Franzese et al., 2006]. Because there is no evidence for a LGM to Holocene change in the end-member sediment compositions [Franzese et al., 2006], the lower glacial 87Sr/86Sr of this mixture must be explained by a reduced input of Agulhas-derived material and/or increased contributions from the other sediment sources with lower 87Sr/86Sr. The flux of material originating from the southern source was much greater during glacial periods [Bayon et al., 2003; Franzese et al., 2006; Kuhn and Diekmann, 2002; Walter et al., 2000], and this must explain at least part of the 87Sr/86Sr signal in these sediments.
 On the basis of the results of our prior provenance and flux study using cores from the South Atlantic, Cape Basin, and a few cores beneath the Agulhas Current [Franzese et al., 2006], the glacial end-member contribution from the southwest Atlantic was about 3.4 times greater than during the Holocene. These results also show that the local input remained fairly constant, while the continental input into the Agulhas Current at the LGM was about twice its Holocene value. Provenance and flux data show that the increased flux of southwest Atlantic material was transported clear across the Atlantic Ocean, into the region south of Africa. The increased flux of Agulhas material, however, appears to be deposited close to its source, indicating a weakened Agulhas Current flow [Franzese et al., 2006]. In fact, end-member modeling and mass balance calculations show decreased LGM fluxes of Agulhas-derived material downstream beneath the Agulhas Current [Franzese et al., 2006].
 The isotope data we present here do not include trace element concentrations (Table 1), and in most cases the age models are not of high enough resolution to resolve differences between Holocene and LGM sedimentation rates. Furthermore, we have made no quantitative estimates of sediment redistribution (e.g., on the basis of fluxes of 230Th). For these reasons, we cannot make the same kind of quantitative arguments as in our previous study [Franzese et al., 2006]. We do, however, have enough information to make a semiquantitative reconstruction of the LGM to Holocene change for a few sediment cores. Four cores from the Agulhas Plateau have 87Sr/86Sr, sedimentation rates, and compositional data for both the Holocene and LGM, which we used to calculate the mass accumulation rates (MAR, g cm−2 ka−1) of bulk sediment and of the terrigenous component (Table 2).
Table 2. Holocene and LGM Mass Accumulation Rates of Agulhas Plateau Cores
Percentage of calcium carbonate measured by coulometry at LDEO.
Percentage of terrigenous is (100 − %CaCO3 − %Opal).
Bulk MAR = (Sedimentation rate) (DBD), where dry bulk density (DBD) = 0.000053(%Carb)2 + 0.00093(%Carb) + 0.3 [Froelich et al., 1991].
MARTerr = (Bulk MAR)(%Terrigenous)/100.
ΔMARAgulhas = LGM MARAgulhas − Holocene MARAgulhas. Using the three sedimentary end-members of Franzese et al. , Holocene MARTerr = Holocene MARLocal + Holocene MARSW.Atlantic + Holocene MARAgulhas and LGM MARTerr = LGM MARLocal + LGM MARSW.Atlantic + LGM MARAgulhas, where MARLocal, MARSW.Atlantic, and MARAgulhas are the MARs of terrigenous material derived from each of the end-members. Assuming that the local terrigenous flux was the same for both time slices, and LGM MARSW.Atlantic = 3.4(Holocene MARSW.Atlantic) [Franzese et al., 2006], this reduces to the following: LGM MARAgulhas − Holocene MARAgulhas = LGM MARTerr − Holocene MARTerr − 2.4(Holocene MARSW.Atlantic).
 Assuming that the local terrigenous flux was the same for both time slices, and the southwest Atlantic was a factor of 3.4 higher during the LGM [Franzese et al., 2006], we can calculate the difference between the Holocene and LGM fluxes of Agulhas-derived material to these four cores using the following mass balance equation: LGM MARAgulhas − Holocene MARAgulhas = LGM MARTerr − Holocene MARTerr − 2.4(Holocene MARSW.Atlantic), where MARTerr is the MAR of terrigenous material and MARAgulhas and MARSW.Atlantic are the MARs of Agulhas-derived and southwest Atlantic–derived material, respectively. For all four of these cores, LGM MARAgulhas − Holocene MARAgulhas < 0 (Table 2). In other words, the LGM accumulation of Agulhas-derived material on the Agulhas Plateau was smaller than during the Holocene. On the basis of their southerly location, we would expect the Agulhas Plateau cores to be the most significantly affected by the increased glacial flux of southwest Atlantic material. Yet even in these cores we can see clear evidence for a reduced glacial Agulhas input.
 On the basis of evidence that the glacial Agulhas Current traveled along its modern trajectory [Winter and Martin, 1990; this study], we infer that the reason for the lower glacial 87Sr/86Sr in the downstream Agulhas Current and retroflection region was a smaller sediment transport capacity, i.e., a weaker Agulhas Current flow. This is consistent not only with a smaller flux of terrigenous material with an east African provenance but also with a reduced temperature contrast between the west and east coasts of the Indian Ocean and more subtropical (rather than tropical) fossil faunal assemblages [Be and Duplessy, 1976; Hutson, 1980]. A smaller flux of material with high 87Sr/86Sr carried by the Agulhas as described above, mixing with a larger flux of material with low 87Sr/86Sr [Bayon et al., 2003; Franzese et al., 2006; Kuhn and Diekmann, 2002; Walter et al., 2000], could easily explain the lower 87Sr/86Sr of sediments deposited in the Agulhas Retroflection region during the LGM. Our main conclusion is therefore that a general weakening of the Agulhas Current system, rather than a displacement of the retroflection, led to the reduced Agulhas Leakage observed for the LGM.
 The lack of evidence for a glacial repositioning of the Agulhas Retroflection has implications for glacial reconstructions of the STF and our understanding of the mechanistic links between the latitude of the STF and Agulhas Leakage. If the mean location of the Agulhas Retroflection remained fixed near its modern position, as our data suggest, then the STF cannot have been significantly north of its modern latitude in the region directly south of the Agulhas Retroflection (∼15°E to 20°E). The reduced glacial Agulhas Leakage that has been widely inferred from an assortment of paleoceanographic proxies [Berger and Wefer, 1996; Charles and Morley, 1988; Cortese and Abelmann, 2002; Flores et al., 1999; Peeters et al., 2004; Pether, 1994; Rau et al., 2002; Verardo, 1995] was not caused by a displacement or pinching of the retroflection due to a northward shift in the STF (or the zonal wind profile) as suggested by numerical models [de Ruijter and Boudra, 1985; Matano, 1996].
 The results of this study also challenge our understanding of the link between Agulhas Current strength and the Agulhas Leakage. It has been suggested that given a stable wind field, a weaker Agulhas Current allows greater westward penetration of Agulhas waters and increased leakage into the Atlantic Ocean basin, while a stronger Agulhas Current forces an earlier retroflection, thereby reducing the Agulhas Leakage into the Atlantic Ocean basin [de Ruijter et al., 2004; Lutjeharms and van Ballegooyen, 1984; Ou and de Ruijter, 1986]. The glacial scenario, however, appears to be one with a weaker Agulhas Current, reduced Agulhas Leakage, and no significant change in the location of retroflection. Our assumption of a weakened glacial Agulhas Current based on changes in terrigenous 87Sr/86Sr is consistent with the reduced temperature contrast between the west and east coasts of the Indian Ocean and a smaller flux of terrigenous material with an east African provenance and more subtropical (rather than tropical) fossil faunal assemblages [Be and Duplessy, 1976; Franzese et al., 2006; Hutson, 1980]. It is, however, inconsistent with the hypothesis that the reduced glacial Agulhas Leakage was caused by a strengthened Agulhas Current.
 The apparent discrepancy between the paleoceanographic data and numerical models must be due to problems with either the data or the models or both. The paleoceanographic data span a large number of cores and a wide variety of proxies. None of the proxies are exclusively controlled by surface current velocity or hydrography: biologically mediated proxies respond to changes in temperature, salinity, productivity, and growth rate and changes in their own microenvironment; proxies measured on terrigenous (or bulk) sediment can be affected by the continental provenance, transport processes, particle flux, and syndepositional or postdepositional sediment redistribution. The good agreement between these fundamentally different proxy methods gives us confidence that the interpretations of a weaker Agulhas Current, reduced Agulhas Leakage, and no significant change in the location of retroflection at the LGM are robust. The inability of the models to adequately reconstruct a scenario that is consistent with the glacial data implies that the system is not yet adequately understood. It may be that the modern relationships do not hold during glacial periods, or maybe we do not have a firm grasp of the complex relationships between forcings controlling the modern Agulhas Current system. Clearly, further work must be done in order to fully understand the glacial mode of the Agulhas Current system and its role in the global THC.
 The 87Sr/86Sr ratios of the terrigenous detrital component of Holocene sediments clearly show that in the region south of Africa, values ≥0.720 trace the mean integrated flow path of the modern Agulhas Current through the retroflection and into the Agulhas Return Current (Figure 3). The terrigenous 87Sr/86Sr data for the LGM do not indicate any significant change in the location of the Agulhas Retroflection compared to its modern position. Relatively low 87Sr/86Sr values northeast of the Agulhas Plateau combined with the constraint of continued Agulhas Leakage argue against a significant northward displacement of the LGM retroflection upstream of 36°S. The consistency of the LGM 87Sr/86Sr from the Agulhas Plateau over a range of water depths corresponding to AABW and CPDW indicates that the Agulhas Retroflection must have been west of the Agulhas Plateau.
 The simplest and most plausible explanation for the pattern of terrigenous 87Sr/86Sr of sediments deposited in the Agulhas Retroflection region during the LGM is that the Agulhas Retroflection remained positioned there during the LGM. The lower than Holocene 87Sr/86Sr is explained by a weaker Agulhas Current system with less capacity to carry material downstream (combined with a greater flux of material with low 87Sr/86Sr from distant southwest Atlantic sources). This implies that the past reductions in the Agulhas Leakage were caused by reductions in the Agulhas Current strength, as opposed to changes in the position of the Agulhas Retroflection. This is inconsistent with the hypothesis of an inverse relationship between Agulhas Current strength and the Agulhas Leakage, which has been proposed on the basis of modern observations and supported by several model studies.
 We thank Lloyd Burckle and Ulysses Ninneman for providing stratigraphic information. This work benefited from discussions with Wilhelmus de Ruijter and Nick McCave, as well as comments and suggestions from Wally Broecker, Bob Anderson, and Yair Rosenthal and reviews by Alex Piotrowski and an anonymous reviewer. This study was supported by NSF grants OCE 98-09253 and OCE 00-96427 to S.L.G. and S.R.H., by an NSF Graduate Research Fellowship to A.M.F., and by a grant/cooperative agreement from the National Oceanic and Atmospheric Administration. Samples used in this project were (1) provided by the Lamont-Doherty Earth Observatory Deep-Sea Sample Repository, supported by the NSF (grant OCE 00-02380) and the Office of Naval Research (grant N00014-02-1-0073), and (2) collected aboard RRS Charles Darwin Cruise 154, funded by the NERC (grant NER/A/S/2000/01161, PIs R. Zahn and I. R. Hall).