Unit TD6 (Table 1) has a maximum thickness of 2.40 m and is constituted essentially of calcareous conglomerates, pebbles and limestone gravel within a matrix composed of sand and clay (Pérez-González et al., 2001). TD6 lies 1 m below the Matuyama–Brunhes limit (located at the top of layer TD7) (Parés and Pérez-González, 1999), and has been dated by a combination of ESR and palaeomagnetic techniques to between 780 and 857 ka (Falguères et al., 1999), and by thermoluminescence to > 960 ± 120 ka (Berger et al., 2008). A new palaeomagnetic study indicates a minimum age of 936 ka for the archaeo-paleontological record of TD6 (Parés et al., 2013). TD6 stands out for the presence of human remains classified as Homo antecessor (Carbonell et al., 1995) with evidence of cannibalism (Carbonell et al., 2010 and Fernández-Jalvo et al., 1996).

This study involved analysis of a sample of 570 pieces from sublevels TD6-1 and 2 (excavations carried out between 1994 and 1996 and between 2003 and 2006). To date the total number of pieces recovered at TD6 is 999 (Table 3). The lithic industry of TD6 has been classified as Mode 1, though it displays certain features which point to developments at the oldest Early Pleistocene sites in southern Europe, such as raw material management (particularly the preferential use of Cretaceous chert for instruments made with flakes) and the use of varied, more complex knapping strategies (Carbonell and Rodríguez, 2006, Carbonell et al., 1999 and Ollé et al., 2013).

Raw material management is characterised by the use of mostly Neogene chert. The secondary raw materials at the site are Cretaceous chert, limestone, quartzite, quartz and sandstone. All these raw materials are local and appear within a maximum distance of five kilometres (Fig. 2, Table 2).

Instruments made from pebbles (0.1%) are scarce, while those made from medium-sized thick flakes are more plentiful. A high percentage of hammerstones and manuports (12.5%) have been recovered. The knapping strategies documented in Unit TD6 are characterised by the use of orthogonal angles. This orthogonal knapping could have been conditioned by the morphology of the Palaeozoic materials (quartzite and sandstone) that are on the Arlanzón T4 terrace, which is the nearest to Gran Dolina (Fig. 3-1). A unique feature in TD6 is the presence of large Neogene chert cores (up to 26 cm in length) with a lack of intensive reduction (Fig. 3-6). The hominins of TD6 produced 23.4 flakes per kilogram from chert and an average of 18 flakes from other materials. Among the retouched flakes there is a significant presence of denticulates, convex dihedrals (scrapers), and concave dihedrals (notches).

Unit TD10 (Table 1) is divided into four lithostratigraphic subunits (TD10.4 to TD10.1, from bottom to the top), with a maximum thickness of 2.50 m. At the time when TD10 was formed the entrance to the cave was in the form of a large rock shelter caused by the collapse of the roof. ESR combined with uranium-series dating give ages of 337 ± 29 ka for the upper part of TD10-1; 379 ± 57 ka at the bottom of TD10-1 and 418 ± 63 ka in the upper part of TD10-2 (Falguères et al., 1999). Thermoluminescence dating has given ages of between 244 ± 26 ka for TD10-2 and 430 ± 59 ka for TD10-3 (Berger et al., 2008).

For this study we analysed a sample of the lithic repertoires of TD10-1 (914 pieces) and TD10-2 (1853 pieces). Up to now a total of 31,680 artefacts have been found in TD10-1 and TD10-2, although 9366 were indeterminable due to preservation problems (Table 3). The lithic industry of the upper levels of TD10 (TD10-1 and TD10-2) has been classified as transitional between Modes 2 and 3 (Ollé et al., 2013). This lithic industry is characterised by a rational, intensive management of the best-quality raw materials, and a tendency towards standardisation of small pieces. The most commonly used raw material is Neogene chert (Table 2). The use of chert increases in TD10-2, reaching as much as 95.7% of the total. TD10 produced 170 flakes per kilogram from chert and an average of 38 flakes from other materials.

In TD10, the chaîne opératoire is fragmented, because part of the knapping was done away from the site. Flake production processes are characterized by the low percentage of cores and their introduction to the site at intermediate knapping phases. The Levallois and Discoid methods were developed almost exclusively on chert (Fig. 3-9). The Discoid knapping method stands out among other operative processes used at the site. The Levallois method has been documented but in small numbers and a variety of forms (Ollé et al., 2013 and Terradillos-Bernal and Rodríguez, 2012).

Few instruments made from limestone, quartzite or sandstone pebbles have been recovered in TD10-1 and TD10-2, and all were made with low to medium quality materials (Fig. 3-5). Denticulated dihedrals and convex continuous dihedrals (scrapers) predominate (Fig. 4-8–10 and 12).

The Neogene chert belongs to the Late Miocene (Pineda, 1997), and it is found in the form of large blocks that outcropped following the erosion of the Astaracian marls and marly limestone in which they were contained (Ollé et al., 2013).

Neogene chert is highly heterogeneous in its reaction to knapping (Table 6, Fig. 2-4). The heterogeneity consists of two main areas: the interior, which features little crystallization and contains geodes; and the exterior (between 1 and 10 cm), which features a very thin, homogeneously crystallized grain. The outside material is not excessively hard, whereas the interior requires greater knapping intensity due to the lack of the transmission of force. The relative softness of this material means it can be knapped with small hammers.

Cretaceous chert belongs to the Turonian-Lower Santonian (Pineda, 1997). This is a fossiliferous chert with a microcrystalline wackestone structure (Ollé et al., 2013).

Cretaceous chert has a medium to high suitability for knapping but this is limited by its small size, polyhedral and thick morphology, and the presence of some irregularities and fissures (Table 6, Fig. 2-2). It is a material that requires the use of low-middle force at the initial stages of knapping and middle-intense force during the advanced reduction phases, especially if the angles are abrupt. Proportionally large hammers and narrow percussion platforms have to be used to prevent overshots.

Utrillas quartzite is a Cretaceous material (Albian). The tenacious rocks (compact quartzarenite), or the very tenacious (metaquartzite, orthoquartzite) have been generically labelled quartzite (Ollé et al., 2013 and Pineda, 1997).

Utrillas facies quartzite (Table 6) is a very abundant material and has good qualities for knapping (Fig. 2-1). Its advantages are that it is easy to control the products obtained, it results in a limited number of accidents and it produces regular and very hard edges, but its spherical morphology, its hardness and the fissures present in approximately 70% of the pieces are its primary disadvantages. Its morphology requires the use of large, dense hammers and sharp strikes by the knapper, especially during the phases in which the size is reduced and percussion angles increase.

The Arlanzón River provides Palaeozoic materials, from La Demanda ridge (quartzite, sandstone and quartz). Among the different varieties of quartzite from the Arlanzón river terraces (Table 6, Fig. 2-6), we analysed the ones with the best qualities for knapping. This raw material represents 3% of the pebbles (according to the analysis of a sample of 800 pebbles) in the Arlanzón river terrace area. They are large, irregular pebbles with an average length of 10 cm. The grain is thin, the structure is homogeneous and they tend to lack fissures or fractures. The experimental analyses have allowed us to know that their suitability for knapping is only ordinary because the cortex impedes the correct transmission of force, resulting in short, quadrangular products.

Sandstone represents 21% of the pebbles in the different terraces of the River Arlanzón (Table 6, Fig. 2-5). These are medium-sized to large pebbles with an average length of 9.1 cm and tend to be flat. Our experiments show that sandstone is a soft material with a homogeneous structure and thin to medium-sized grain. It features a considerable number of frontal fissures that determine the knapping processes and cause numerous fractures. The reduction of this material normally requires the use of a wide platform, which generates massive products. The flakes resulting from this raw material are very irregular, have sinuous edges, little cutting capacity and are very fragile. This is a very suitable material for the production of large, forceful, flat tools.

Quartz represents 13% of the pebbles in the terraces of the River Arlanzón (Table 6, Fig. 2-7). These pebbles are small and thick with a thin to medium-sized grain and have some irregularities and fissures that cross the three base axes. The fissures cause fractured, hinged flakes (many of them with Siret fractures), which have limited potential for use.

Limestone was procured from the karst, from the Cretaceous substratum of the Sierra de Atapuerca (Ollé et al., 2013). Limestone is the most abundant and the geographically closest raw material in the Sierra de Atapuerca (Table 6, Fig. 2-3). It appears in thick and irregular blocks distributed both inside and outside the caves. Its properties make it a poor material for knapping and it is highly heterogeneous.

We carried out eight experiments using the Discoid method (well represented in Middle Pleistocene assemblages). When it is used on Neogene chert the Discoid method allows a large number of flakes with homogeneous morphologies to be extracted using wide percussion platforms and limited force. In the final stages of exploitation, principally with quartzite and Cretaceous chert, more accidents occur with this method, most commonly fractures (often Siret fractures) and hinged flakes (Fig. 3-8 and 11).

The Levallois method was used in 15 experiments in its different variants, in three of them combined with the Discoid method. This combination is rare in the archaeological assemblages studied, but has a great qualitative value. The Levallois method is considerably efficient (high/middle) when it is used on Neogene chert (six experiments) and is even more efficient when combined with the Discoid method. On Neogene chert, the best results with Levallois method were obtained from thick flake blanks. On Cretaceous chert (four experiments), the Levallois method seems to be a relatively inefficient strategy due to two basic reasons: the limited size of the blanks, which makes prehension difficult, and the presence of at least one fissure. The products that typically tend to be related to this method were not generated (Fig. 3-8).

When we used Utrillas quartzite (two experiments) the Levallois method was perfectly adapted to the homogeneous qualities of the material. The quartzite from the River Arlanzón (three experiments) yielded numerous hinged flakes and the initial morphometry of the blanks was worse suited to the Levallois method.

We conducted nine experiments using the unifacial/bifacial massive unipolar method or using the centripetal method (Fig. 3-2–4). In experiments with Neogene chert we were able to select large, very thick bases with wide, non-cortical initial platforms. This allowed us to obtain large, thick flakes from the first reduction phase, almost with a predetermined morphology if the edges and angles were well controlled. This is a very opportunistic and quick method, suitable for use with large fragments of Neogene chert.

Six experiments were carried out with bipolar knapping on anvil on Neogene chert (n = 3) and quartz (n = 3). None of the experiments improved upon the results produced with other techniques and methods.

In Gran Dolina TD6 and TD10 instruments made on pebbles are scarce and all were made of low to medium quality materials like limestone or quartzite of scanty quality because they are short cycle tools, which only need a heavy weight. We verified experimentally that the making of pebble tools is very simple, affecting only a small proportion of the edge and closely connected to the use of weight as an active factor.

We verified that the selection of the initial raw materials for the configuration of handaxes and cleavers (documented in Galería) gave priority to morphology over quality, having selected flat sandstone pebbles and big Neogene chert flakes.

In the case of small flake instruments the production is characterized by its simplicity and speed (between 10 seconds and 2 minutes) (Fig. 4-7, 11, and 14). Neogene chert yields the largest formats and the sharpest edges. Cretaceous chert presents more regular and resistant edges, but requires a greater investment of force (Table 6). The small size of Cretaceous chert pieces does not allow numerous reactivation phases to be carried out. Retouch is easy on the quartzite and sandstone of the Arlanzón terraces, although the resulting flake edges are duller and less regular. The rectification of knapping accidents is more complex with these materials.

The flake production documented in Unit TD6 of Gran Dolina consists of a small but very significant assemblage of quartzite cores and large chert cores. Through experimentation, we verified that the sphericity of the pebbles found in terrace T4 of the River Arlanzón (the terrace nearest to the site) influenced the development of massive, orthogonal knapping products. Palaeozoic materials from terrace T4 of the River Arlanzón were used in TD6 because of its proximity to the site but these hominins had little ability and/or interest to modify the initial formats. The experiments with quartzite, quartz and sandstone from the Arlanzón River present the most fissures and the most spherical shapes, which results in knapping with very abrupt angles (Fig. 3-3 and 4). In general, it seems that the hominins at these sites did not place much importance on the initial selection of the stones or on the modification of the initial blanks.

In TD6 the hominins abandoned the exploitation of large blocks of Neogene chert in the most optimal phase for the extraction of flakes. Our experimental knapping shows that these cores were abandoned after an initial series of cleaning removals (Fig. 3-6–7). In light of this fact, two complementary hypotheses can be put forward: •

these cores were part of a reserve of raw materials;

In the development of the Levallois and Discoid methods the initial format (thick and trapezoidal) of most of the blocks of chert is very important. In Middle Pleistocene assemblages the Discoid method was the most suitable for the raw materials used in the production of small flakes. Its use in association with the Levallois method results in more fluid knapping during the last phases of reduction.

In the experimental program an average of 22 flakes were obtained per kilogram from chert (with a maximum of 82 and a minimum of 9) and 17 from Palaeozoic materials (quartzite, sandstone and quartz), with a maximum of 38 and a minimum of 10. It can be seen that the production from Palaeozoic materials is more homogeneous (because it offers fewer possibilities). If we compare it with the archaeological record, we note that TD6 yielded 23.4 products per kilogram of chert, Galería 37 and TD10 170; whereas in TD6, other raw materials generated 18 flakes per kilogram in Galería 9.6 and in TD10 38.

A considerable difference was found in the proportion of flakes between the most ancient levels of Galería and the most modern. Why is there such a big difference in the proportion of flakes in these levels? There are two possible explanations for these differences: •

there is a progression in the increase in the size of the flakes, so there is an inversely proportional relationship between the size of the flakes and the quantity of flakes;

The experimental program on quartzite and limestone verified that early hominins collected blanks that enabled them to quickly produce heavy, blunt instruments through a very short knapping process. So, in TD6 fractured cores and cores with noticeable knapping accidents were used as well as fractured hammerstones, fragments and limestone pebbles. In TD6 and TD10, medium quality raw materials were selected to shape instruments from pebbles.

Galería contains a significant presence of large handaxes and cleavers (Fig. 4-1). In order to produce these flat and large tool flakes the hominins selected flat pebbles intensifying the acquisition of sandstone (21% among pebble instruments and 9.1% among tool flakes). We verified through diverse experiments that sandstone is a raw material that offers a scanty resistance and allows the fluid application of the knapping with a soft hammer. The production of these instruments made us stricter in selecting the blanks: they had to be flat, of scanty resistance to the fracture and without fissures. In the experiments in which we did not use soft hammers we generated thick and asymmetric large cutting tools, with a triangular thick sagittal cross-section. Instruments made with the soft hammer technique are flatter and more symmetrical (from the sagittal edge), there are more hinged removals (but less marked), and extractions are narrow, very flat and rectangular. This technique has proven to be effective only with Neogene chert and sandstone, due to the initial morphology of the blanks and their reduced resistance to fracture. The great energy investment that we have used in the experiments and that past hominins used in the production of these instruments must have been related to the planning of activities, which would imply the use of multifunctional instruments.

The flake instruments in the archaeological record generally show evidence of expeditious shaping. The retouched flakes made of Cretaceous chert have regular and durable cutting edges. Little striking intensity was needed to retouch the Neogene chert flakes, as we corroborated in our experimental program. The hominins at these sites knew how to correctly assess the qualities of Cretaceous chert to produce small instruments, although they did not attain a great degree of control in knapping this raw material. Therefore, the proportion of the use of this stone is higher among flakes and small implements (primarily in TD10-2 and TD6) (Table 7).