We characterized the effect of large-scale (>20 mm) soil physical structure on the age and recalcitrance of soil organic carbon (SOC) in upper (A) and lower (B) horizons of grassland soils from California's Central Valley. The radiocarbon content of SOC from surfaces and interiors of large-scale soil structural units (“peds”) was measured in order to characterize the spatial distribution of soil C pools with distinct residence times. The radiocarbon content of CO2 released following sieving was used to identify the C that is readily respired upon physical disturbance of soil structure. We found the longest SOC residence times in the interiors of peds from subsurface B horizons, where limited bioturbation leads to stable large-scale structure. The radiocarbon value of this interior SOC (Δ14C = −555‰) indicates that this pool has been protected from decomposition for thousands of years. Similarly ancient C (Δ14C = −596‰) was released upon physical disruption of subsurface B horizons from a similar soil, indicating that this SOC was old, but chemically labile. With cultivation, the C released upon physical disruption of B horizons was even older (Δ14C = −812‰) than in the uncultivated soil. In uncultivated A horizons, which are subject to continuous bioturbation, large-scale structure resulted in contrasting SOC pools only in the surface horizon, where “bomb” C effects are strong. A horizon incubations also suggested effects of smaller-scale structure. Loss of the labile SOC that is physically protected by large-scale structure contributes to the rapid reduction in natural soil C inventories following cultivation.
 It is widely recognized that soil organic matter consists of an array of compounds of differing turnover times [e.g., Trumbore, 1993], but controls on these turnover times are not fully understood. Efforts to identify C pools with distinct turnover times have applied chemical and physical separation techniques to field samples, often neglecting the natural physical structure present in most soils. At the scale of particles or small aggregates, there is evidence that carbon-mineral association can have a role in soil C dynamics [Basile-Doelsch et al., 2005; Masiello et al., 2004; Rasmussen et al., 2005; Torn et al., 1997], and it is well known that that small-scale (<2 mm) aggregation of soil particles in agricultural and surface soils leads to slower turnover of soil organic C [Six et al., 2002]. However, large-scale (>20 mm) soil structural units (referred to here as “peds”) are commonly observed in situ, within soil profiles. To our knowledge, the effect of these peds on rates of C cycling has not been investigated. Well-developed peds offer a clear, easily defined starting point for investigating large-scale structural effects on C cycling. Although they have the potential to create distinct spatial distribution of soil C pools with respect to both age and recalcitrance, little is known about how this scale of structure affects C cycling.
 We undertook two studies that tested (1) whether soil ped structure leads to a distinct spatial distribution of soil C pools; (2) whether peds are effective in protecting otherwise easily degraded organic matter from decomposition for long periods of time; and (3) if, upon physical disturbance, much of this protected C is readily mineralized. We used radiocarbon analysis of peds from a mature and strongly structured grassland soil to characterize the spatial distribution of SOC pools. A complementary incubation study tracked radiocarbon released as CO2 upon physical destruction of soil structure in samples from a similar soil, from both cultivated and uncultivated sites [Wang et al., 1999]. Together, these two studies link the spatial variation in ped-associated residence times prior to disturbance, with the susceptibility of SOC to decomposition following simple physical disturbance.
 This work also has important practical implications for studies of soil C dynamics. For example, soil samples are commonly sieved to obtain a homogenized sample for laboratory analysis [e.g., Fang et al., 2005]. If large-scale soil structure affects soil C turnover rates, this practice will lead to a loss of resolution among spatially distinct C pools, and may result in the release of structurally protected but chemically labile C that is then not accounted for in sample analysis.
 Soil structure is an integral part of field observations of soils [National Soil Survey Center, 2002] because it is associated with many key processes of soil formation, including biological activity, hydrologic conditions over the lifetime of the landform, weathering of primary minerals to form clays, and downward clay transport. In many soils, the combined activity of microorganisms, plant roots and burrowing animals mixes high concentrations of organic matter near the surface with low to moderate (2–20%) amounts of clay, forming granular structural units on the order of 1 to 10 mm in size. In some cases, larger-scale, secondary blocky structure is also apparent. With increasing depth (∼0.5 m) in many temperate soils over 104 years in age, substantial increases in clay (up to 30% or more) and limited biological mixing, coincide with the presence of blocks, prisms, or columns that are ∼10 to 100 mm in size.
 Field observations of soils often include an assessment of the “grade” of soil structure (i.e., “weak” to “strong”), referring to the distinctness of in situ peds and their coherence when disturbed. The development of strong subsurface peds has been most clearly associated with the size distribution of mineral particles (e.g., increased clay with depth) and the degree and rate of wetting and drying by rain or groundwater [Moniz and Buol, 1982; Southard and Buol, 1988; van der Graaf, 1978]. Ped faces are thought to begin as desiccation cracks, and these cracks are more likely to repeatedly form with slow cycles of wetting and drying [Moniz and Buol, 1982; van der Graaf, 1978; White, 1966]. Slow application of drying stress causes the soil to crack along pre-existing planes of weakness, leading to more distinct and durable peds in poorly drained soils [Southard and Buol, 1988]. As peds develop, the organic C they contain may be isolated from sources of oxygen and water, and thus from decomposition, even if this C is relatively “labile.” As a result, interior SOC might be expected to easily decompose upon disturbance of ped structures.
 With time, developing ped structures are reinforced by alteration of their exteriors; for example, ped surfaces may become more weathered, and coated with oriented clays [Southard and Buol, 1988]. The development of large-scale structure also drives preferential flow and solute transport [Vervoort et al., 1999], and it is a common observation that roots preferentially grow in the cracks between large peds. As a result, the ongoing flux of relatively young C through the soil will come in contact preferentially with ped faces that are likely to be rich in clays and other reactive minerals. This suggests that SOC on ped faces will be younger than in ped interiors.
 While most SOC is concentrated near the land surface in the soil zone of granular to weak blocky structure, an estimated one third of the global soil C pool is stored below the top meter [Jobbagy and Jackson, 2000]. In many environments, this subsurface C cycles much more slowly than surface C, turning over primarily on millennial rather than decadal timescales [Baisden et al., 2002b; Gaudinski et al., 2000]. It is possible that development of soil ped structure contributes to this pattern. For these reasons, understanding subsurface structural controls on soil C is critical to characterizing terrestrial organic C dynamics.
3.1. Soil Sampling
3.1.1. Ped Study
 In 2001, we excavated intact peds (n = 3 per horizon) from a San Joaquin series soil (fine-loamy, mixed, thermic Abruptic Durixeralf) in the eastern San Joaquin Valley of California (38°9′N, 121°8′W; Figure 1). The taxonomic classification of this soil indicates a subsurface horizon enriched in pedogenic clay, referred to as a “Bt” or “argillic” horizon, which places it in the Alfisol soil order (“-alf” in the classification above [Soil Survey Staff, 1999]), as well as a horizon cemented by pedogenic silica (“Duri-” indicating a “duripan” [Soil Survey Staff, 1999]). This well-developed soil is derived from granodioritic alluvium, originating from the Sierra Nevada Mountains in central California (Riverbank Formation). It is found on minimally eroded terraces and fans, so that the soil age reflects the depositional age of the starting alluvium (∼105 years old [Harden, 1982, and references therein]). The climate is Mediterranean (hot, dry summers and cool, wet winters), with MAT of 16°C and MAP of 300 mm. The flora consists of annual grasses and forbs, with widely dispersed oaks.
 The profile was excavated and described according to standard methods [National Soil Survey Center, 2002], and peds were collected from the A1, AB and Bt3 horizons (three of seven identified horizons; see Table 1). The A1 horizon was distinguished by its abundant fine roots and darker color, indicating substantial organic C accumulation, as well as smaller structural units compared to the horizon below it (A2, Table 1). The AB horizon is a transitional horizon with visible organic C (dark color), but with weaker structural units and slightly higher clay content (18%) than the overlying A horizons (Table 1). The Bt3 horizon has a high clay content (40%), redder colors, and strong angular blocky structure (Table 1). Upon removal from the profile, peds were marked to indicate orientation (top/bottom), and placed in sample bags on ice for transport back to the lab, where they were immediately freeze-dried for approximately 3 days.
Table 1. Soil Description of the San Joaquin Soil Used for the Ped Carbon Studya
 Soil samples were collected from an upland grassland soil (Fallbrook series; fine-loamy, mixed, thermic Typic Haploxeralf) found in the central Sierra foothills (36°43′N, 119°17′W) and from an adjacent lemon orchard soil (converted to agriculture 22 years before sampling in 1994) [Wang et al., 1999]. The mean annual precipitation and temperature are 310 mm and 18°C, respectively, reflecting the somewhat more southern latitude than that of the San Joaquin profile. The parent material is granodioritic bedrock. The Fallbrook soils are located on convex hillslopes subject to erosion, and their approximate age can be quantified as a residence time [e.g., Heimsath et al., 1997; Yoo et al., 2005a]. Given erosion rates for the Sierra Nevada of ∼1 cm/100 years [Janada and Croft, 1967; Riebe et al., 2001], the residence time of these soils is on the order of 104 years. Like the San Joaquin, the Fallbrook soil has a subsurface horizon with high pedogenic clay content. Given local climate conditions, the presence of Bt horizons in the Fallbrook soil also indicates substantial time for soil development (∼104 to 105 yr) [Harden, 1987]. Thus the Fallbrook and San Joaquin soils possess key similarities for purposes of this paper.
3.2. Ped Sample Analysis
 Ped exteriors were sampled by using dental tools to carefully scrape undamaged surfaces, which in many cases were visibly darker and smoother than ped interiors. To ensure representative sampling, entire faces were scraped for each replicate. Replicate samples were collected from three peds. Ped interiors were also sampled in triplicate, after physically splitting peds open using a hammer and chisel. Samples were ground by hand using a corundum mortar and pestle. Visible roots were removed using tweezers. Subsamples were then weighed into tin capsules for total C and N analysis (∼100 mg soil per tin). Additional subsamples (∼500 mg soil) were weighed into 6 mm Vycor tubes that were then loaded into 9-mm Vycor tubes with 1 g Cu, 1 g CuO, and a strip of Ag foil. Tubes were evacuated and sealed, combusted at 875°C for 3 hours, and held at 650°C for 2 hours, before further cooling [Minagawa et al., 1984]. The resulting CO2 gas was cryogenically purified and manometrically quantified prior to Δ14C analysis at Lawrence Livermore National Laboratory (LLNL). Results for %C, %N, C/N and Δ14C were compared using a one-way analysis of variance (JMP 4.0 software), followed by pairwise comparison of means using the Tukey-Kramer Honestly Significant Difference test (α = 0.05).
 For the incubation study, field moist soil samples from 0–30 cm (pooled A horizons) and 30–70 cm (pooled B horizons) were collected from the grassland and orchard sites. Samples were homogenized and sieved to remove the coarse fraction (>2 mm), and visible roots were removed by hand. Separate incubations were conducted for respiration rates and radiocarbon, for each of the four treatments (grassland A and B, orchard A and B). In all incubations, temperature was held at 15°C and soil moisture was held at 60% of water holding capacity.
 For determination of soil respiration rates, 100 g of soil was incubated in quart jars for up to 45 days. Respiration rates were calculated at regular intervals by monitoring the CO2 concentration in the headspace of the incubation jars using a LiCor 6200 infrared gas analyzer. The orchard soil B horizon treatment was replicated (n = 4), and variation in observed respiration rates was <10% for samples taken at similar timepoints (within 2 hours).
 For radiocarbon analysis of respired CO2, 1–2 kg of soil was incubated in 1-gallon containers. Cumulative CO2 respired over 2- to 5-week periods was trapped and then cryogenically purified prior to 14C analysis at LLNL. For three treatments (grassland A, orchard A and B), two replicate incubations were sampled and successfully analyzed for radiocarbon in the first 20-day interval. For other time intervals, radiocarbon values for respired CO2 were evaluated once per treatment.
3.4. Δ14C Analysis
 At LLNL's Center for Accelerator Mass Spectrometry, CO2 was converted to graphite using a H2 reduction method [Vogel et al., 1984] and loaded into targets for Δ14C analysis. Δ14C was calculated as
where Asn is the activity of the sample normalized to its δ13C value (approximated from Baisden et al. [2002b] for ped samples) and Aabs is the activity of the standard (0‰), which represents approximately the 14C/12C of the preindustrial atmosphere. Using this Δ notation, positive values indicate incorporation of recent atmospheric CO2 enriched in 14C produced by nuclear weapons testing (“bomb” C). The long-term accuracy and precision (1 σ) of this technique on modern C (Δ14C > 0‰) is better than 9‰ [Vogel et al., 1984].
 The sharp increase in atmospheric 14C levels between 1955 and 1964, and subsequent decline following the moratorium on atmospheric testing, has proved to be a powerful tracer of recent C dynamics [Randerson et al., 2002; Trumbore, 2006, 2000]. This known variation in atmospheric 14C may be used to interpret a positive Δ14C value as reflecting a mean residence time for an actively cycling C pool [e.g., Baisden et al., 2002b; Gaudinski et al., 2000; Trumbore, 2000; Wang et al., 1996]. However, even simple models may not provide a unique solution for mean residence time using a single Δ14C value and associated timepoint, and interpretation of solutions may be hindered by other uncertainties (e.g., contribution of a “passive” or other discrete, differently cycling pool). Here our intent is to assess whether large-scale soil structure defines C pools that are clearly cycling differently within representative soil profiles. This degree of difference is implied by Δ14C values suggesting at least order-of-magnitude differences in residence time (e.g., a high positive value indicating decadal-scale cycling versus a very negative value indicating millennial-scale cycling).
4. Results and Discussion
4.1. Ped Analysis
4.1.1. Surface (A1) Horizon
 Large-scale structure was present in the surface horizons (A1, A2) of the San Joaquin soil (Figure 1), but was more weakly developed than in the subsurface (Table 1). Secondary structure was also present in these peds: the larger subangular blocks (∼100 mm in size) broke relatively easily into smaller subangular blocks (∼20 mm) and granules (∼10 mm), along faces where small roots were present and coloration was variable. In accordance with the presence of this intermediate-scale, secondary structure, our sampling at the scale of primary ped structure did not reveal significant differences in %C and %N between ped surfaces and interiors (Figures 2a and 2b). However, the C:N ratio was slightly higher on ped exteriors versus interiors (Figure 2c), suggesting that organic matter inside peds is slightly more decomposed [Baisden et al., 2002a]. In these C-rich, A1 horizon peds, both the surface and interior C are modern (Δ14C > 0; Figure 2d), with surfaces significantly (α = 0.01) more enriched in 14C (Δ14C = 49 ± 12‰) than interiors (12 ± 16‰). This difference may indicate focused delivery of very recently fixed organic C to ped surfaces by downward moving water and roots. The effect of bomb C will be most pronounced at this near-surface depth.
 These results are generally consistent with the hypothesis that the structure in the A1 horizon is dynamic, and is disrupted by physical and biological processes over decadal timescales. Gopher and ground squirrel burrowing, with accompanying soil movement, is a pervasive process in this region [Arkley and Brown, 1954], and complete mixing of A horizons by these organisms can occur on 100-year or shorter timescales [Johnson, 1990; Yoo et al., 2005b]. As a result, physical protection in structural units in A horizons can be expected to have a relatively short-term effect (101–102 years) on C cycling. Similar timescales have been demonstrated for C sequestration by small-scale soil structure in agricultural systems [e.g., Kong et al., 2005], and for preferential flow paths for SOC transport in forest soils [Hagedorn and Bundt, 2002]. This timescale of mixing suggests that in this horizon, the difference in radiocarbon values between surfaces and interiors is a function of the pronounced effect of bomb C on C inputs at this near-surface depth.
4.1.2. AB Horizon
 In this intermediate horizon, %C and %N were not significantly different on ped surfaces versus interiors (Figures 2a and 2b), although mean %C values were lower in ped interiors, resulting in slightly lower C:N ratios in interiors versus surfaces (Figure 2c). This suggests that organic matter is slightly more decomposed inside peds. However, radiocarbon results did not differ for surfaces versus interiors (Δ14C = −77 ± 5 and −66 ± 19‰, respectively). The similarity of radiocarbon values for AB surfaces and interiors suggests that this horizon is subject to bioturbation and mixing, as suggested for the A1 horizon, but with a different C input rate and composition. Both the surfaces and interiors in AB peds were more depleted in 14C than surfaces and interiors in A1 peds (49 ± 12 and 12 ± 16‰, respectively; Figure 2d). A sample from the base of this horizon was slightly more depleted in 14C than both AB surfaces and AB interiors (Figure 2). These trends are consistent with the decrease in radiocarbon values commonly observed with soil depth [e.g., Baisden et al., 2002a], and indicate that C inputs to this horizon may be slower and older than those of the A1 horizon. As a result, the effect of bomb C at this depth will be less pronounced than in the A1 horizon. Combined with century-scale physical mixing by burrowing organisms [Johnson, 1990; Yoo et al., 2005b], this could result in ped surfaces and interiors with similar observed radiocarbon values.
 Minimal bioturbation below the AB horizon can be expected to result in more pronounced differences between ped surfaces and interiors with respect to C cycling. This is borne out by results for the Bt3 horizon, below.
4.1.3. Bt3 Horizon
 Ped structure was most pronounced in the Bt3 horizon, where clay content was also highest (40%). These peds have distinct faces, edges and corners (strong angular blocky structure; Table 1), and are easily separated from one another when removed from the profile (Figure 1). In contrast with the horizons above, both %C and %N in this horizon were significantly higher on the ped exteriors (0.28% and 0.05%, respectively) than in the interiors (0.07% and 0.02%, respectively; Figures 2a and 2b), suggesting limited mixing or transfer of organics from surfaces to interiors.
 C:N ratios in the Bt3 horizon were markedly lower in ped interiors (3.7) versus exteriors (6.1), suggesting that slow decomposition of a relatively isolated organic pool may be occurring inside peds (Figure 2c). Δ14C values were much lower in ped interiors (−555‰) versus surfaces (−234‰), indicating millennial versus century-scale turnover times, respectively. We infer that the slowness of organic C cycling within this subsurface zone is a function of physical protection of organic matter within peds. It is likely that this physical protection has been enhanced over time as soil structure has developed, resulting in an increasing mean residence time of subsurface organic C with time.
 The Bt3 horizon lies well below any evidence of bioturbation and burrowing. The radiocarbon data, in addition to revealing sharp spatial differences in C cycling, also support the concept that structural units in these clay-rich B horizons have residence times on the order of 103 years. There is a profound difference in structural effects on C cycling between A horizons (“biomantles” as proposed by Johnson ) and underlying Bt horizons, where abiotic effects dominate development of soil structure.
 The surfaces and interiors of peds in the Bt3 horizon had lower C:N ratios than in A1 and AB horizons, and this horizon was significantly more depleted in 14C compared to those nearer the surface (Figure 2d). This increase in organic C residence times and degree of decomposition with depth is similar to other soils in the region (vertical lines in Figure 2d [Baisden et al., 2002b]), and implies that transport-associated aging leads to the decline in 14C. The marked spatial differences that we have observed in ped surface C versus interior C occur within this overall depth trend.
 The base of the Bt3 horizon rests on an indurated Si-cemented hardpan (a “duripan”). At the interface is a pronounced dark layer (Munsell color 10YR 3/2) ∼5 mm thick, where many fine roots were observed. This layer had 4 to 5 times as much C and N as vertical faces, but its C:N ratio and Δ14C value were similar to other Bt3 surfaces. It appears that this large accumulation of subsurface C and N is due primarily to preferential flow around peds, which may either directly transport dissolved organics, or support increase root or microbial biomass that in turn served to concentrate C and N. The hardpan interrupts downward water movement, resulting in accumulation of downward moving material.
 Clearly, soil ped structure controls both preferred flow paths and the proliferation of roots and microbial biomass in the subsurface, and the intimate connection between roots, microbial biomass, and hydrology in turn controls the distribution of soil C [Bundt et al., 2001a, 2001b]. On the basis of this study, it is not possible to discern whether the C on ped faces has arrived via roots or dissolved flow, but the implication is that both should be considered, and both are controlled by structure.
 Radiocarbon signatures of the initial CO2 pulses from the uncultivated grassland A (−88‰, −147‰) and B (−596‰) horizons (Figure 3) were considerably more depleted in 14C than SOC in whole samples (125 ± 30‰ and −356 ± 8‰, respectively; data from Wang et al. ). To consider whether these pulses reflect release of older C due to disturbance of soil structure, we compare our results to field measurements of respiration from this soil by Wang et al.  over the course of a year (1995). We interpret these field measurements as the undisturbed “control” for comparison to the incubation results. Soil respiration is a mix of root and organic-matter-derived (microbial) CO2, and this “control” reflects a ∼50% contribution of CO2 by root respiration during the height of the growing season. However, root contributions were minimal during the dormant season, and soil respiration was primarily organic-matter-derived. Despite this seasonal variation in sources, in situ soil respiration was consistently enriched in 14C (100–200‰). Overall, the range and seasonal variation of these field observations generally suggest that the microbial-derived respiration is not highly depleted in 14C [Wang et al., 2000]. A comparison of radiocarbon values for total soil organic C, field respired C, and C respired in the incubation (Table 2) indicates that the pulse of 14C-depleted CO2 that we observed in the first 20 days of the incubation for both the A and B horizons represents physically protected C that is both old and chemically labile, and does not contribute significantly to in situ soil respiration.
Table 2. Comparison of Radiocarbon Values for the Fallbrook Soil: Soil Organic C, CO2 Efflux From Undisturbed Soil, and CO2 Released During Laboratory Incubations
Total C respired during the incubation experiments (39 to 45 days).
 The initial Δ14C value of −596‰ released from the B horizons of the Fallbrook grassland soil is comparable to the value observed for the Bt3 ped interior in the San Joaquin soil (−555‰; Figure 2d). Thus the millennial-scale residence time of C released in the incubation (1) corresponds to the residence time of SOC preferentially found within well-developed peds, and (2) occurs within B horizons where bioturbation is minimal. This is consistent with the idea that physical protection of organic C in the subsurface is a function of large-scale soil structure.
 The ped sampling in the A and AB horizons of the San Joaquin soil (Figure 2a) did not yield the contrasting residence times observed by incubating the A horizons of the Fallbrook grassland soil (Figure 3), i.e., the depletion of 14C in physically protected C (−118‰) relative to the starting sample (+125‰). This difference may be a function of differences between these two soil profiles; for example, variable structure within A horizons generally, or between the A horizons of these two particular soils. However, it is possible that smaller structural units may dictate most physical partitioning of SOC in A horizons where roots and burrowing animals are abundant. This idea is consistent with many previous studies showing smaller-scale (<2 mm) structural protection of organic C in surface horizons [e.g., Six et al., 2002], and merits further investigation.
 During incubation of the sieved orchard A horizons, the initial CO2 pulse was enriched in 14C (−10‰, +11‰; Figure 3) compared to total C (−41 ± 21‰). In the orchard B horizons, the initial CO2 pulse was more depleted in 14C (−716‰, −908‰) compared to total C (−251 ± 21‰). These results suggest that previous disturbance of this soil (conversion to orchard) led to rapid (∼20 years) incorporation of modern C into soil structure in A but not B horizons. Smaller-scale structure and decadal to century-scale C partitioning processes characterize A horizons, while larger-scale structure dictates millennial-scale C partitioning processes in subsurface B horizons.
 The Δ14C value and fractional amount of CO2 released from the B horizons of the orchard soil (−812‰; 17 mg C g−1C) were lower than those observed in the uncultivated soil (−596‰; 31 mg C g−1C). These exceedingly low Δ14C values imply that this labile C has persisted in a biologically unavailable state for thousands of years (apparent 14C ages are 13 and 7 ky, respectively). The comparison between land use histories suggests that disturbance-induced decomposition, and perhaps the loss of dissolved C through application of irrigation water with cultivation, has limited physical protection to a smaller, more recalcitrant or well-protected pool of subsurface organic C. The total soil C storage was reduced by 26% in the orchard following conversion to agriculture [Wang et al., 1999], yet this soil respired 40% less of its C than the grassland soil in the incubation (Figure 3; normalized to total C). All of this evidence suggests that labile C had been lost with cultivation, as has been observed in many studies [see Amundson, 2001, and references therein; Miller et al., 2004; Preston et al., 1994].
 In the years since radiocarbon analysis first revealed that soil C must consist of multiple C pools of differing residence times [O'Brien and Stout, 1978; Trumbore, 1993], there has been a concerted effort to physically and chemically identify those pools. In this work, we tested the hypothesis that development of large-scale soil structure spatially partitions SOC pools with distinct residence times. The results demonstrate the importance of large-scale soil structure in physical protection of subsurface SOC over millennial timescales. Because physically protected, old C is not necessarily biochemically recalcitrant, physical disturbance has the potential to quickly release a portion of this otherwise slowly cycling C pool.
 Separation techniques intended to identify soil C pools of differing lability or residence times must account for the physical disturbance inherent in the very act of collecting or preparing a sample. Our work suggests that the nature of the soil structure observed in the field should dictate how samples are collected and analyzed. For example, the use of a 5-cm core to collect a field sample will fail to capture and preserve in-situ relationships when peds are 10 and 20 cm in size.
 The development of large-scale soil structure occurs by virtue of limits on physical mixing of soil particles (bioturbation, physical erosion). Large-scale soil structure drives root architecture, microbial activity, and associated C cycling. Ped development also defines physical pathways of water mediated transport, focusing and limiting movement of both dissolved C and weathering products, and dictating the extent to which mineral surfaces are available to interact with soil C and N. We have observed pronounced variation in soil C residence times at spatial scales defined by readily identifiable, large soil structural units commonly present in well-developed soils. We suggest that key aspects of soil C cycling such as organo-mineral associations [e.g., Masiello et al., 2004] may be focused at structurally defined boundaries, and these boundaries should be the subject of future study.
 We thank Carrie Masiello, John Southon, Brian Franz, and other staff at LLNL for assisting with target preparation and graphitization, and for generally making these 14C measurements possible. We thank Leonard Root and Greenleaf Farms for access to their property. Funding for this work was provided by the California Agricultural Experiment Station and the Western Regional Center for Global Environmental Change. Additional support for W. T. B. was provided by the New Zealand Foundation for Research, Science and Technology.