As a matter of fact, fossil wood is mostly encountered as isolated pieces of secondary xylem. Primary xylem, pith and periderm are sometimes found in connection, but such situation is exceptional, and for the following “wood” will be understood as “secondary xylem”.

Fossil wood taxonomy was born while descriptive obsession and fixist views dominated palaeontology. Until at least the mid-twentieth century, palaeobotany's main goal was to make a census of fossil diversity and to order it within a system which was designed for extant plants. “Genera” and “species” of fossil wood were described; binomials were used, which more or less implicitly hypothesised that the lowest taxonomical level fitted with that of the species recognised by the neobotanists. Such was satisfying as long as species were considered as morpho-anatomically determined units. The neobotanical and palaeobotanical species concepts, however, diverged as fast as the species concept integrated biological and genetical considerations.

This prompted palaeobotanists to have an appendix to the ICBN to accommodate the taxonomic challenges posed by fossil plants (Bateman and Hilton, 2009), and to distinguish, as an artificial taxon, the form-genus. The form-genus was “maintained for classifying fossil specimens that lack diagnostic characteristics indicative of natural affinity but which for practical reasons need to be provided with binary names” and conceived as a taxon “within which species may be recognized” (Lanjouw et al., 1956). At the same time, the somewhat parallel concept of organ-genus was proposed which, in 2000, was eventually merged with that of form-genus into the new concept of morphogenus, an evolution of which Bateman and Hilton (2009) remarkably discussed the ins and outs.

As for tracheidoxyls, the important fact is that the ICBN establishes a taxonomical unit, at the genus rank, within which species can be recognized. The choice of this ranking is purely arbitrary, however, even if it can be understood by the convenience of binomials and by the importance given formerly to the identification of fossil specimens at family level. It is clear, now that family diagnosis has shifted from morpho-anatomy to molecular phylogeny, that the former belief according which almost every isolated plant fossil could be assigned to a modern family is an utopia, rooted in times predating evolution theory. Even for extant conifers, a character-based xylological system of their woods does not fit with a phylogenetical system (see, for e.g., the similarity between the wood of some Podocarpus and that of some Taxodiaceae, Marguerier and Woltz, 1977). Morphotaxa being arbitrary ranked entities, they are not comparable. Although they may represent similar disparities, they do not necessarily encompass the same diversity (i.e. that of corresponding extinct biological species). Two morphogenera (resp. species) may correspond to Linnean (biological) taxa of different hierarchical order.

To sum up, a wood morphogenus groups fossil wood specimens fitting with a diagnosis, and morphospecies can be distinguished. These taxonomical units do not fit a priori with the genus and species level of neobotanical classification, morphogenus and morphospecies being hierarchically ordered but unranked taxa. Morphotaxa do not necessarily univocally fit with taxa designed for extant plants.

Fossils fitting with the Araucarioxylon wood-type, as described above, are encountered worldwide, from the Carboniferous (Devonian?) to the Present (Frentzen, 1931 and Giraud, 1991). More than four hundred morphospecies have been described (partial reviews in Frentzen, 1931, Jeffrey, 1913, Schultze-Motel, 1962 and Schultze-Motel, 1964). Confronted to this unusually long time-span and numerous descriptions, several authors have tried to split or subdivide the morphogenus. Morgenroth (1883) and Felix (1886) proposed to limit the use of Araucarioxylon to the Mesozoic and Cainozoic material, and to group the Palaeozoic fossils within a taxon called Dadoxylon, an earlier nomenclatural synonym of Araucarioxylon. Caspary (1887) advocated the use of two names, Araucarites Göppert for those woods lacking axial parenchyma and Araucariopsis Caspary for the others. Knowlton (1890) choose to assign all the Araucarioxylon-species found once in connection with cordaites pith to the genus Cordaites. Considering that no clear line of demarcation could be drawn between them, Gothan (1905) recommended grouping all woods described as Araucarioxylon Kraus, Araucarites Göppert, Cordaioxylon Felix, Cordaioxylon Grand’Eury, Cordaites and Dadoxylon Unger under the later name. Zalessky (1911) thought that Dadoxylon should be used only for fossils with both primary and secondary xylem, while Araucarioxylon for tracheidoxyls, a choice which was later advocated by Prasad (1986). Seward (1919) introduced the use of trinomials, like Dadoxylon (Araucarioxylon) novae zeelandiae, for fossil woods of the Araucarioxylon-type which could “safely be assigned to the Araucariaceae”, and kept Dadoxylon binomials for wood of more uncertain affinities. Dealing with Palaeozoic taxa only, Frentzen (1931) distributed these woods within two groups according to if the radial pitting completely covered the tracheid or not. More recently, Lepekhina (1972) and Doubinger and Marguerier (1979) have proposed other taxonomical schemes.

Closely related genera were described as segregates from the main “Araucarioxylon” group, like Dammaroxylon Schultze-Motel and Simplicioxylon Andreanzsky. The former was distinguished (Schultze-Motel, 1966) on the basis of the occurrence of “Randzellen” (better translated as “marginal spaces” than as “marginal cells”), the latter as its ray cells are sometimes pointed (Philippe, 1993). The taxonomical relevance of these xylological features could be discussed, but might be of interest given the paucity of other characters. Planoxylon Stopes, basically woods of the Araucarioxylon type with Abietineentüpfelung, is definitely well enough circumscribed in both time (Mesozoic) and space (southern Gondwana) to be worth distinguishing (Philippe and Hayes, 2010). This is also recognised for Australoxylon Marguerier.

Because of sedimentological sorting, wood is rarely encountered in anatomical connection with other types of palaeobotanical remains. The Araucarioxylon-type of wood was observed connected or associated to Caytoniales, Glossopteridales, Cordaitales, Cycadales, Voltziales, Ullmanniales, Araucariaceae, Cheirolepidiaceae, etc., but such observations are rare and never evidenced a peculiar anatomical feature which could allow a subdivision of this set of woods.

At the morphospecies level several taxonomical reappraisals were attempted (Frentzen, 1931, Giraud, 1991 and Lepekhina and Yatsenko-Khmelevsky, 1966), but such are getting more and more challenging given the multiplication of new names, regularly established with limited knowledge of relevant literature, based on poorly preserved specimens, or with poorly illustrated protologues. Often a new specimen is compared only with those species which have been described from the same geological stage. Various sets of fossil wood with an Araucarioxylon-type of anatomy were investigated with numerical taxonomy, including multivariate analyses, but described grouping or trends are mainly due to database incompleteness (Falcon-Lang and Cantrill, 2001, Giraud, 1991 and Li, 1988). Booi (2010), applying several multivariate techniques to a large collection (ca 250 specimens) of araucarioid wood from the Palaeozoic of Sumatra, found a very varied but homogenously coherent group. Even for the wood of modern Araucariaceae statistical approach has a low discriminative power (Ilic, 1995). Because they were closely similar to already described wood morphospecies, but slightly different, some fossils were used as type for “sister-species” and named accordingly, like Dadoxylon parafuronii Boureau & Koeniguer, Dadoxylon subrhodeanum Grand’Eury or Dadoxylon pseudoparenchymatosum Gothan. Such is common in neobotany, but less understandable for morphospecies, all the more since precaution is usually the main reason advocated for the founding of these “new species”.

The reasons which are most of the time put forward while a new Araucarioxylon morphospecies is described are, in increasing frequency order: geographical origin; age; and xylological singularity. In a wholly character-based approach, age and geographical origin of a fossil are irrelevant to its taxonomy (Bateman and Hilton, 2009 and Frentzen, 1931). Singularity relatively to xylologically, most similar morphospecies is usually judged from what is known of anatomical diversity among modern Araucariaceae. Such is typical of confusion between morphospecies and species, is an induction on fossil taxonomy that cannot be justified by uniformitarism, and underestimates the long-known intra-individual and intra-specific xylological variability documented by extant species (Patton, 1927, Pool, 1929 and Welch, 1927). Particularly significant in this respect is the sentence by Evans (1934) “a distinction between fossil Araucaria and fossil Agathis based solely upon the position of a few bordered pits in a fragmentary tracheid certainly appears quite valueless to any one with but a passing acquaintance of the many tricks which our Agathis australis plays with its pitting”. This sentence was written well before the discovery of the Araucariaceae third extant genus, Wollemia, but the wood of the later does not differ significantly from the family pattern (Heady et al., 2002). It is a strong hypothesis to suppose that the systematical distribution of wood features, as well as the xylological variability of taxa, remained unchanged through times, all the more since the number of extinct taxa is probably much higher than that of extant taxa.

Be the xylological variability of modern Araucariaceae what it may, fossil taxa boundaries must be based on discrepancies within the morphological space described by the fossils. Could yet described fossil woods with an Araucarioxylon-type of anatomy be distributed within a morphological space, no clear boundaries would appear through what would be a continuous cloud as far as most if not all the xylological variability of the fossil species can be observed within the extant Araucaria araucana (including rootwood and wood of traumatic area) (Gondran et al., 1997).

Focusing on one relatively short time interval and limited area, or dealing with a limited number of specimens, one might have the impression that fossil wood specimens with Araucarioxylon-type of anatomy can be distributed in discrete xylological units. This is probably because taphonomical filter drastically sorts among the wood fragments getting into the sedimentary system. Only a low percentage of biological tree species and only some parts of their woody body are recorded in a geological stage fossil record. When only a few specimens are known, their xylological disparity has a high probability to be discrete. With about 400 species yet described displaying the Araucarioxylon-type of anatomy and with an estimated number of three thousand published specimens at most for a 350 My time interval, the sampling is obviously meagre. Also induced by the taphonomical process is what could be named the pseudo-variability, i.e. the apparent variability which is due to poor preservation or erroneous interpretation. This is a common problem in palaeoxylology, and does not spare Araucarioxylon (Savidge, 2007).

Whereas by other fossil wood morphogenera (e.g. Xenoxylon Gothan), clear xylological discontinuities are observed, by what is usually called Araucarioxylon nothing similar is noted while applying a character-based and bottoms-up approach.

This point already inspired countless pages. For a long time, palaeobotanists have named their taxa on the basis of supposed relationships with extant taxa. If a fossil wood displayed araucarian radial tracheid pitting and araucarioid cross-fields, i.e. a wood anatomy similar to that of modern Araucariaceae, it had to be named Araucar-ites (Göppert, 1850 and Tuzson, 1911) or Araucario-xylon (Kraus, 1870). As a matter of fact, such approach was still recently advocated (Yang et al., 2000). However, this way of naming fossils (identification sensu Bateman and Hilton, 2009) is clearly inspired by a fixist view of palaeontology, according which every fossil had to fit within a modern taxon, and is in complete contradiction with both the reality of evolution and the ICBN. Practically, the name Araucarioxylon does not imply per se any systematic relationship with the Araucariaceae, nor a special type of anatomy. Etymology cannot compensate for a poor diagnosis or a poorly preserved type specimen.

Binomials have been used as a “short-hand” representation of a particular character suite (Bateman and Hilton, 2009). Therefore names like Araucarioxylon biseriatum, A. parenchymatosum, or A. crasseradiatum were given. Again these names are just labels on morphotaxa which are defined by a diagnosis and, if this is incomplete, by a type (provided it is well enough preserved, micron-scale details having to be observed).

Up to the 1980s, most authors used the taxonomical and nomenclatural framework proposed by traditional literature (see, for e.g., Kraus, 1870 and Kräusel, 1949). Warnings about the invalidity or illegitimacy of the Araucarioxylon name were published (Gothan, 1904 and Vogellehner, 1964), however, and the number of new Araucarioxylon binomials regularly decrease after 1920, becoming less frequent than the new Dadoxylon names (Fig. 2). Curiously, after the 1980s, the proportion of new Araucarioxylon binomials rose again (vs. Dadoxylon), and during the last decennia the former name completely dominated. Be that a matter of fashion, or a late consequence of Felix's proposal correlated to a decrease in the naming of Palaeozoic tracheidoxyls, anyhow Dadoxylon and Araucarioxylon are both invalid names. Now that it is clear that several extinct organisms have no extant equivalent, and that priority rules also the use of morphotaxa names, fossil wood nomenclature must be completely revisited (Philippe and Bamford, 2008).

At least 16 morphogenera were used for fossil wood having (or thought to have) an anatomy of the Araucarioxylon type, as defined above: Agathoxylon Hartig, Araucariopsis Caspary, Araucarioxylon Kraus in Schimper, Araucarites Endlicher sensu Göppert, Araucaroxylon (as an orthographic variant, vide Fliche), Baieroxylon Greguss, Cedroxylon Kraus in Schimper, Cordaioxylon, Cordaites, Cormaraucarioxylon Lignier, Dadoxylon Endlicher, Dammaroxylon Schultze-Motel, Palaeoxylon Brongniart, Peuce Lindley & Hutton, Pinites Witham, Platyspiroxylon Greguss, Pseudagathoxylon Greguss, Simplicioxylon Andreanzsky. In a previous nomenclatural study (Philippe, 1993), it was concluded that Agathoxylon Hartig is the earliest validly published name that can be used to name fossil woods with an Araucarioxylon-type anatomy. This proposition was either accepted, e.g. (Crisafulli et al., 2009, De Witt et al., 2009, Gnaedinger and Herbst, 2009, Ottone and Medina, 1995, Poole and Mirzaie Ataabadi, 2006, Salunkhe and Yagyani, 2006, Valenzuela et al., 1998 and Zamuner and Falaschi, 2005), questioned (Falcon-Lang and Cantrill, 2001) or ignored, e.g. (Ash, 2003, Lucas et al., 2010, Morgans-Bell and McIlroy, 2005 and Noll et al., 2005).

Of course, one may want to submit a proposal for the conservation of Araucarioxylon, but such would be difficult to defend given that Dadoxylon is used almost as frequently, that Araucarioxylon was neotypified by Andrews with a syntype (A. carbonaceum (Witham) Kraus) based on a poor sample (Witham of Lartington, 1833), and that even recently Araucarioxylon was used inconsistently (compare e.g. Duan, 2000, and Wang et al., 2000). It is probably wiser to use Agathoxylon.

It was exposed previously that the xylological variability of modern Araucariaceae is not relevant to the circumscription of fossil wood morphospecies. Palaeoxylological taxonomy must be based on character analysis and fossil wood disparity must be distributed among artificial taxa, ordered within a two-level only hierarchy.

The 428 species included in the database were taxonomically reappraised. For most of them (310) the protologue or a type could be accessed. Unfortunately, no decision could be made for 118 species, either as the original material was too poor or as we did not manage to get the original description. Among the 428 species of the database, 223 fit with the Araucarioxylon-type of wood as defined above. The distribution in time and space of this 223 species is very uneven: 12.3% were described from India, 10.5% from Germany and 9.1% from Russia, whereas three huge potential reservoirs of fossil wood diversity (Karoo formations, Madagascar and western USA) account as a whole for only 7% of the recorded species; 17% of the species are from the Permian, whereas only 7% from the Jurassic, which is however longer. The ratio of the number of species described vs. the duration is varying from stage to stage (Table 1), with clearly artificial peaks in the Permian and the Cretaceous. If, as supposed by Forey et al. (2004), about 50% of the existing fossil species are yet recorded, a conservative estimate of circa 800 species could be advanced.

Given the intrinsically small number of characters and of character states in the mature wood of the Araucarioxylon-type (Greguss, 1955 and Jeffrey, 1913), and given the variability observed in fossil specimens, four hundred or more species is questioning. This forces to face a merger-splitter dilemma. A merger attitude would point out that, as no major xylological discontinuity can be observed among the fossil woods of the Araucarioxylon-type from at least the Early Permian (Wang et al., 2000) to the present, no morphospecies but one should be recognized. As the earliest wood described fitting with the Araucarioxylon-type, Pinites brandlingii Lindley & Hutton should probably be the basionym of that species (ICBN, art. 11.4). However, the corresponding new combination (Agathoxylon brandlingii) is not proposed here as the type material has not yet been reviewed.

Conversely, a splitter would argue that, as xylological variability can be distributed within several arbitrary classes, a large number of species can be recognised. For example, if four classes are made for radial pitting (mostly uniseriate, mostly biseriate, etc.), four classes for rings (lacking, weak, marked with gradual transition, marked with abrupt transition), four classes for rays (low, average, high, extra high), four classes for cross-fields (whatever the numerical values) and two classes for axial parenchyma (present vs. absent), then it is theoretically possible to recognize 512 species; and it would not be difficult to have much more.

The 1956 version of the code, by Lanjouw et al., 1956, used to define the form-species as entities “which for practical reasons need to be provided with binary names”. What should morphospecies be recognised for? In the literature, wood morphospecies have been used for biostratigraphy, palaeobiogeography, palaeoecology, phylogeny, biodiversity analysis, etc. For all these applications, the wood species must have a limited extension in both time and space. Whatever the solution adopted to the merger/splitter dilemma, this condition of limited extension will not be fulfilled for woods of the Araucarioxylon-type.

Moreover, most anatomical features used up to now to distinguish species among this group are environmentally controlled. The seriation of radial pits and the number of cross-field pits are pro parte functions of tracheid width, itself a function of water availability. The ray height is, pro parte similarly, a function of the distance to the pith, and thus of trunk diameter. The wood of the Brazilian Araucaria angustifolia (Bertol.) Kuntze may have growth-rings or not depending of its provenance. These anatomical features are also genetically controlled, but this part of the determinism can only be deciphered with statistical analysis of numerous samples (Ilic, 1995). Such an approach is probably not realistically applicable to fossil wood.

Binomials, which are memory-friendly to most botanists, are conveniently used to handle fossil woods. In the first times of palaeoxylology, when only a few specimens were studied, within what was then known as Araucarioxylon, anatomical variation was discrete. It was thus completely scientifically sound to establish morphospecies. Now that documented variability is more complete, there is apparently no reason anymore to distinguish species within this group of woods. As groundbreaking as it may appear, this conclusion is just the logical consequence of research progresses. There is a lot to learn about palaeobiogeography, palaeoecology or phylogeny using morphogenera only (Philippe et al., 2004 and Philippe et al., 2009). The latitudinal oscillation of the Agathoxylon/Xenoxylon Exclusion Line in Eurasia during the Mesozoic is potentially an interesting terrestrial climate proxy (Oh et al., in press). The conclusion that only one species should be recognized within Agathoxylon does not throw the interest of palaeoxylology back into question, but rather urges the invention of postdescriptive palaeoxylology.