In 1989, reviewing the available radiocarbon determinations from Australia (including Tasmania), New Guinea and the nearer Melanesian Islands, I observed that there was no appreciable difference in the oldest radiocarbon ages from this region, from north to south and east to west, with the exception of the centre of Australia where the oldest desert date was then c. 22,000 years ago (Allen 1989). A general antiquity, based on radiocarbon, of 35-40,000 years for the initial colonization of Australia had not been extended during the previous decade, and I concluded that current evidence gave little reason to go beyond a date of c. 40,000 years. However, within a few months Roberts, Jones & Smith (1990a) had published thermoluminescence (TL) dates for the Northern Territory site of Malakunanja II suggesting that it was first occupied by humans between 50,000 and 60,000 years ago. Nine TL dates in stratigraphic order span the period 200+1,300 years (KTL-156) from the very top of the site to 107,000+21,000 years (KTL-163) at a depth of 4.58 m, about 2 m below the point where artefacts cease. The vital dates towards the bottom of the archaeological deposit are 45,000+9000 years (KTL-164) from a depth of 2.30-2.36 m; 52,000+11,000 years (KTL-158) at 2.41-2.54 m; and 61,000+13,000 years (KTL-162) at 2.54-2.59 m. This latter date corresponds with the lowest artefact recovered during the 1988 re-excavation of the site (Roberts et al. 1990a: 154). There is a marked peak in artefact density from 2.3 to 2.5 m and thus the date of 45,000+9,000 years is seen as a minimal date for human occupation; a linear least-squares regression of the nine dates suggests that 50,000 years is a conservative age for the sediments surrounding the lowest artefacts (Roberts et al. 1990b:127). Note that the errors are not standard deviations in the usual sense, but rather are 'total uncertainties' (see Roberts et al. 1990b: 126; Roberts in press). As Roberts et al. (1990b: 126) note, while these wide error margins pertain, greater confidence can be given to the central tendencies of luminescence dates if multiple, closely spaced samples disclose a pattern of steadily increasing age with depth. While true, at the same time there is a logical constraint to this general argument of stratigraphic superpositioning; it will still work if there is a constant error in the data, such as an inappropriate dose rate in the equation. By itself depth-age correlation is no demonstration of real age, but merely of consistency between samples.

This increase in the accepted antiquity for humans in Australia of between 25% and 50% not unexpectedly drew questions, comments and criticisms (Bowdler 1990; 1991; Frankel 1990; Hiscock 1990) and responses (Roberts et al. 1990b; 1990c). The questions followed two main lines, firstly concerning the technique and especially the large standard deviations associated with the TL dates, and secondly about the association of the human artefacts and the dated sand sediments. The possibility of artefacts moving down into older deposits, raised by both Hiscock and Bowdler, was rejected by Roberts et al. on the grounds that the mixing of artefacts and previously deposited sands in a 'kick zone' would re-set the TL 'clock' of these sediments. Other taphonomic processes which might transport artefacts into older sands have not been considered.

The case of Roberts et al. has been recently been significantly strengthened by their announcement of a similar age for the basal deposits of a second Arnhem Land site, Nauwalabila I, 65-70 km south of Malakunanja II. This site contains ‘securely stratified' artefacts in a rubble base below the sand deposits dated by the related but different luminescence technique, optically-stimulated luminescence (OSL) (Jones 1993; 114; Roberts et al. 1993; Roberts et al. in press). At Nauwalabila I a sequence of five OSL dates are also in stratigraphic order The three oldest samples are 30,000+2400 years (OxODK166) from 1.70-1.75 m depth below surface; 53,400+5400 years (OxODK168) from 2.28-2.40 m; and 60,300+6,700 years (OxODK169) from 2.85-3.01 m. This latter date is below both the rubble layer and the lowest artefacts, while the date of 53,400+5400 years dates the sands immediately above the rubble layer.

The central issue is whether Malakunanja II and Nauwalabila I are really >15,000 years older than any other known Australian site as these dates imply. Luminescence dates measure calendrical years and for that part of the radiocarbon range for which we can calibrate radiocarbon determinations against other dating techniques, uncalibrated radiocarbon determinations mainly underestimate calendrical years. Stuiver et al. (1991: 10) suggest this underestimation is c. 2000 at 14,000 years ago. Mazaud et al. (1991) propose a maximum underestimation of 3000 years between 18,000 years ago and 40,000 years ago and a negligible difference between 45,000 years ago and 50,000 years ago. Bard et al. (1993) indicate that a determination of 18,000 radiocarbon years represents almost 22,000 calendar years. Stuiver & Reimer (1993) use this last date as the oldest in their most recent calibration program. In western NSW, Bell (1991: 48) compared four paired radiocarbon determinations and thermoluminescence dates for separate hearths each c. 30,000 years old, where the TL dates were between 3500 and 5100 years older than radiocarbon determinations. However, substantial comparative sequences of radiocarbon determinations and dates produced by alternative radiometric techniques for the crucial period between 20,000 and 40,000 radiocarbon years are not yet available from anywhere in the world.

A further problem with radiocarbon dating is sample contamination either in nature, by the downward percolation of humic acids or colloidal carbon after burial, or during recovery or processing after excavation. Because an organic sample has lost so much of its radioactive carbon component by the time it is 25-30,000 years old, contamination by even small amounts of modern carbon will cause the radiocarbon determination of an old sample to underestimate real age. Aitken (1990: 85-6) has calculated that the addition of 1% of modern carbon to a 34,000-year-old sample will result in a radiocarbon determination which is too recent by 4000 years, and after this point the magnitude of the error climbs alarmingly. An infinitely old sample contaminated in this way will yield an apparent age of only 38,000 years.

Roberts et al. (1990b: 125-7; 1990c: 95-0) and Jones (1993: 113) have consistently referred to this contamination problem to suggest ‘that the oldest-accepted radiocarbon ages for human occupation in Australia of 35,000 to 40,000 should be regarded as only minimum ages with an unknown upper bound' (Roberts et al. 1990b: 126; 1900c: 95). These authors prefer a contamination explanation for the differences between Australian radiocarbon and luminescence dates to one involving the underestimation of calendrical years by the radiocarbon method. They point out that calibration as currently estimated is insufficient to convert a radiocarbon determination of 40,000 years into a calendrical date of 50,000 years (Roberts et al. 1990c: 95) although as discussed, the point remains to be demonstrated in practice.

The contamination explanation raises archaeological questions. Determinations of c.34-38,000 radiocarbon years are now known from caves, shelters and open sites in various locations from Tasmania to the north coast and offshore islands of New Guinea, and from New South Wales and Queensland to northern and southern Western Australia. These determinations represent bone, shell and charcoal samples taken from diverse environments across 40^o of both latitude and longitude. processed at various laboratories in and outside Australia. If these very similar determinations arise from such a general contamination, it implies that all the Australian sites (and those elsewhere?) with radiocarbon determinations in the range 35-40,000 radiocarbon years are likely 50,000 years old or more, since as little as 0.5% of modern carbon can reduce such a real age to this radiocarbon range.

The problem is that we do not know that various levels of contamination occur so consistently. For example, the contamination proposition predicts that radiocarbon determinations plotted against depth will begin to ‘flatten out' at somewhere less than 30.000 radiocarbon years. Even in undisturbed sites, deeper samples will only be older until this threshold is reached, after which deeper samples will apparently not increase in age much or at all. Yet where good sequences of radiocarbon determinations from undisturbed locations have been recovered, as at several recently excavated southwest Tasmanian sites which date to beyond 30,000 radiocarbon years, this pattern has not been encountered. ft must be added, however, that sequences with sufficient antiquity and enough radiocarbon determinations to test this proposition are still too rare in Australia to generalize on this point. However, we still need to maintain the possibility that the true radiocarbon ages of some or many of these old sites are approximately as measured, without having any way of determining which are which, short of re-excavating the sites and re-dating them by alternative radiometric techniques.

Although Jones (1993: 113) suggests that radiocarbon determinations of 35-40,000 years are close to the method's practical limits, it is generally accepted that accelerator mass spectrometry (AMS) can measure radiocarbon age back to 50,000 years ago. However, up until now the use of the AMS technique has not produced any age determinations associated with human artefacts beyond 38,000 years in Australia or New Guinea. In contrast, John Head (Radiocarbon Laboratory, Australian National Ijniversitv) is producing consistent and reproducible radiocarbon determinations on non-archaeological marine shell from the Great Barrier Reef which approach c. 50,000 radiocarbon years, using conventional (not AMS) techniques. His technique reduces the uptake of atmospheric carbon by preparing the samples in a nitrogen-rich environment.

At Matenkupkum, a New Ireland coastal cave site which may represent a very early incursion by humans (Allen et al. l989: 558), four previous basal radiocarbon determinations on the same species of marine shell clustered between 31,350+550 b.p. (ANU-5469) and 33,300+950 b.p. (ANU-5070). One of the samples, from the same square and depth as ANU-5469, yielded a determination of 32.500+800 b.p. (ANU-5065). A new radiocarbon determination on the same shell species taken from exactly the same depth and in a square adjacent to ANU-5469 and ANU-5065, but prepared using the new technique, yielded a result of 35,410+430 b.p. (ANU-8179). Statistically this latter determination is significantly older than its two close neighbours (Head et al. in prep.) but only extends the younger of these by c. 13% and the older by c. 9%, using the central tendencies, There are thus good reasons for thinking that contamination has been minimised for ANU-8179, and that this site was first occupied c. 35,000 radiocarbon years ago. If Matenkupkum reflects the earliest period of human settlement of New Ireland and we accept the luminescence dates from Arnhem Land at face value, then Malakunanja II and Nauwalabila I were occupied perhaps 20,000 years before New Ireland. Future calibrations of the radiocarbon determinations may well reduce this gap, but at present we can only guess by how much.

As much as I find it improbable that the inland and (at times) semi-arid Arnhem Land escarpment was settled so much earlier than the high biomass tropical coast of New Ireland, arguing the point in our present state of knowledge reduces to matters of opinion and hypothesis rather than fact. While the radiocarbon contamination argument is technically sufficient to reconcile discrepancies between luminescence dates in the Northern Territory and radiocarbon determinations from elsewhere in Pleistocene Greater Australia it remains intellectually unsatisfactory because it potentially discards the baby with the bath water. Some radiocarbon determinations >30,000 b.p. may approximate the real ages of deposits. There is a difference between allowing that sample contamination may have occurred in any sites with radiocarbon determinations of >30,000 b.p. and presuming that it has. So far in Australia we have not demonstrated any instance where a radiocarbon determination has underestimated the age of an archaeological deposit by anything like 10.000 years when dated by a different radiometric technique. This is not to say that this will not occur, it simply says we should not presuppose it, and especially not presuppose it is the norm. Conversely, the absence of AMS or conventional radiocarbon determinations >38,000 b.p. from archaeological deposits anywhere in Greater Australia remains a problem. Too great a gap presently exists between the dating sets.

Confirming the older luminescence dates by duplicating these results in other Australian sites is important and on-going (Rhys Jones pers. comm.). Equally important is developing a better understanding of the relationship between radiocarbon determinations and luminescence dates, particularly by extending the calibrations of radiocarbon further into the Pleistocene. Obtaining paired samples of luminescence dates and radiocarbon determinations from diverse sites where reliable samples for both techniques occur and where human occupation runs through the terminal Pleistocene from c. 40.000 radiocarbon years ago to 10,000 years ago is a major priority. This may be no easy task since deposition rates for the Pleistocene levels of Australian sites are frequently very low and quite different dates may be separated by only a few centimetres of deposit. However some of the southwest Tasmanian sites will allow for unclustered radiocarbon ages over most of this timespan to be compared with a series of luminescence dates taken down each section. Work has begun on this project and is planned to be expanded in the next southern hemisphere summer.

Archaeologically there is little basis for rejecting the Arnhem Land luminescence dates on present evidence. However, accepting them has fundamental implications not only for ideas about water crossings and the initial colonisation of Greater Australia, but also for understanding the nature of subsequent settlement, the multi-regional model of human evolution, modern human behaviour and the spread of early modern humans, prehistoric art, and the human role in the extinction of the Australian megafauna, to note but a few topics. Lacking extinct faunal successions and precise lithic technologies in Australian late Pleistocene sites chronology has always provided the primary analytical framework. Thus, currently we are in the middle of a significant dating revolution.

Acknowledgments. I wish to thank Rhys Jones, Bert Roberts and John Head for unpublished information and discussions. David Frankel and Tim Murray for conversations and comments on an earlier draft of this paper and two anonymous reviewers who will recognise some of their suggestions here which are otherwise unacknowledged.

AITKEN, M.T. 1990. Science-based dating in archaeology. London: Longman.

ALLEN, J. 1989. When did humans first colonize Australia? Search 20(5): 149-55.

ALLEN, J., C. GOSDEN & J.P, WHITE. 1989. Human Pleistocene adaptations in the tropical island Pacific: recent evidence from New Ireland, a Greater Australian outlier. Antiquity 63: 548-61.

BARD, E., M. ARNOLD, R. G. FAIRBANKS & B HAMELIN. 1993. ²³⁰Th-²³⁴U and ¹⁴C ages obtained by mass spectrometry on corals. Radiocarbon 35(1): 191-9.

BELL, W.T 1991. Thermolominescence dates for the Lake Mungo Aboriginal fireplaces and the implications for radiocarbon dating. Archaeometry 33: 43-50.

BOWDLER, S. 1990 50,000 year-old site in Australia - is it really that old. Australian ArchaeoIogy 31: 93.

1991. Some sort of dates at Malakunanja II: a reply to Roberts et al., Australian Archaeology 32: 50-51.

FRANKEL, D. 1991. Time inflation. New Scientist (7 July): 52-3

HEAD, J., J.ALLEN & C.GOSDEN. In preparation. New dates from New Ireland: minimising contamination during the preparation of shell samples for radiocarbon dating.

HISCOCK., P. 1990. How old are the artefacts in Malakunanja II? Archaeology in Oceania 25(31:122-4.

JONES., R. 1993. A continental reconnaissance: some observations concerning the discovery of the Pleistocene archaeology of Australia, in M. Spriggs, DE. Yen, W. Ambrose, R. Jones, A. Thorne & A. Andrews (ed.), A community of culture: the people and prehistory of the Pacific: 97-122. Canberra: Department of Prehistory. Research School of Pacific Studies. Australian National University. Occasional Paper in Prehistory 21.

MAZAUD, A., C. LAJ, E. BARD, M. ARNOLD & E. TRIC. 1991. Geomagnetic field control of ¹⁴C production over the last 80 ky: implications for the radiocarbon time-scale, Geophysical Research Letters 18 (10): 1885-8.

ROBERTS, R. G., R.JONES & M.A. SMITH. 1990a. Thermoluminescence dating of a 50,000 year old human occupation site in northern Australia, Nature 345: l53-6.

1990b. Stratigraphy and statistics at Malakunanja II: reply to Hiscock. Archaeology in Oceania 25(3): 125-9.

1990c. Early dates at Malakunanja II: a reply to Bowdler, Australian Archaeology 31: 94-7.

1993. Optical dating at Deaf Adder Gorge, Northern Territory indicates human occupation between 53,000 and 60.000 years ago. Australian Archaeology 37: 58-9.

ROBERTS, R.G., R JONES. N.A. SPOONER, M.J. READ, A.S. MVRRAY & M.A. SMITH. In press. The human colonisation of Australia: optical dates of 53,000 and 60,000 years bracket human arrival at Deaf Adder Gorge, Northern Territory. Quatuernary Geochronology.

STUIVER, M & P.J. REIMER. 1993. Extended ¹⁴C data base and revised Calib 3.0 ¹⁴C age calibration program, Radiocarbon 35(1): 215-30.

STUIVER, M., T.F.BRASIUNAS, B.BECKER & B. KROMER. 1991. Climate, solar, oceanic and geomagnetic influences on late-glacial and Holocene atmospheric 14C/12C change. Quaternary Research 35: 1-24.