Documented sites of prehistoric collective violence before 3000 BCE

How this record was compiled

A site is included when all four tests are met:

• Before 3000 BCE. The record closes at the conventional boundary of the site's prehistoric period.
• At least three individuals (MNI ≥ 3) from one deposition. Two is an incident; one leaves no context to establish. The individuals must come from the same depositional event, not from a cemetery used over generations — with one stated exception, discussed below.
• Demonstrable perimortal violence. Injury identified by the excavators or the osteological report as perimortal and attributable to violence: cranial fracturing by weapons, embedded projectile points, systematic peri-mortem breakage. Healed trauma establishes a violent life, not a violent death, and does not qualify a site on its own.
• An excavation publication or osteological report as primary source. An overview article may support a row but cannot be its only reference.

The certainty column is about the violence interpretation, not about the dating. High: the excavators' reading is not seriously contested. Medium: the violence is established but its character — massacre, execution, raiding over time — is argued. Low: whether the assemblage records violence at all is contested in the specialist literature.

What was deliberately left out

• Isolated individuals with violent injury. Ötzi, the Porsmose man and the many single burials with embedded points are evidence of violence, not of collective violence. They belong to a separate record and the MNI rule excludes them by design.
• Sites after 3000 BCE. Roaix, Wassenaar and the Bronze Age battlefield in the Tollense valley all fall outside the window. The Tollense valley in particular is the best-documented prehistoric battlefield in Europe and it is roughly 1250 BCE, more than a millennium too late for this record.
• Fortifications and weaponry without human remains. Enclosure ditches, palisades and arrowhead concentrations are read as evidence of conflict in the literature. They are not skeletal evidence and this record is skeletal.
• Cannibalism where no perimortal violence is demonstrated. Processing marks establish what happened to a body after death, not how the death occurred. Herxheim is included because the assemblage is central to the debate, and it is marked low for exactly this reason.
• Estimated or summed totals. The MNI column reproduces the figure the publication gives, in the publication's own terms — "at least 67", "estimated 500 or more". These figures are not comparable and they are not added up anywhere on this page.

Disputed entries

At this record's scale the disputes are not exceptions; they are most of the interesting rows.

Jebel Sahaba was published in 1968 as a single massacre and is still cited that way. The 2021 restudy of the collection found healed alongside unhealed injuries and read the cemetery as the outcome of repeated raiding over a period, not one event. Both readings are in the row.

Nataruk was published in Nature in 2016 as the earliest evidence of inter-group conflict among foragers, and was challenged in the same journal within months: the critique holds that the cranial damage is consistent with post-depositional sediment pressure and that the "bound" posture is not established. It carries low certainty, and its inclusion is not an endorsement of the original reading.

Herxheim divides its own excavation team. One group reads the processed remains as cannibalism; another reads them as a large secondary mortuary rite involving remains brought from elsewhere. No perimortal weapon trauma of the Talheim kind has been demonstrated. It is the least secure row here and the most cited.

Ofnet Cave presents two deposits of selected crania with perimortem injuries and cut marks. Whether this is a massacre followed by curated deposition, or a mortuary practice in which some individuals had died violently, is unresolved — and so is the count. The monograph that fixes the number of crania is not available in any edition this record could open, so the row says what an accessible source demonstrates and leaves the total out rather than repeating a figure it cannot check.

San Juan ante Portam Latinam is the stated exception to the single-deposition rule. It is a rock shelter used for burial over several generations rather than a single event, and it is included because the 2023 restudy argues on the skeletal evidence for a sustained period of conflict rather than accumulated accidents. The row says so, and a reader who applies the single-deposition rule strictly should discount it.

Bergheim and Achenheim both contain deposits of severed upper limbs. Whether these are war trophies taken from captives or a punitive or ritual practice is argued, and the two sites are usually read together.

Sources

Every row cites the study that carries the claim, to a table or a section: Crevecoeur et al. for the Jebel Sahaba restudy; Meyer et al. for Schöneck-Kilianstädten and Halberstadt; Novak et al. for Potočani; Alt et al. for Els Trocs; Fernández-Crespo et al. for San Juan ante Portam Latinam, for Bergheim and Achenheim, and — in the comparative table of their 2023 paper — for the counts at Talheim and Asparn/Schletz; Boulestin et al. and Orschiedt and Haidle for the two sides of Herxheim.

Every row now points to a place in a source — a table, a section or a page range — and not merely to a work. That is the standard this site holds all its datasets to, and for an osteological claim it is the whole point: how many individuals, which injuries, whether perimortem, is exactly where a reader checks whether the row reproduces the report or interprets it.

Completing those references changed four cells, and the changes are worth stating. Talheim and Asparn/Schletz kept their counts of 34 and 67, but they now rest on the comparative table of a 2023 paper that could be opened and read rather than on excavation reports that could not; those reports are still named, marked as context. Achenheim and Bergheim moved to the dates and the pit counts of a 2025 study that re-examined both sites: six individuals in pit 124 at Achenheim, eight in pit 157 at Bergheim, and a range of roughly 4300 to 4150 cal BC for both. Ofnet lost its count altogether. Two rows — Nataruk and Herxheim — rest on publications behind a paywall; their references give the article and its pages, which is what a reader needs to check them, but this record could not open them and does not pretend otherwise.

Supporting reading on this site

• Talheim, 5000 BC: The Pit That Changed the Neolithic Argument — the clearest site in this record, and the reference point against which every contested row here is measured — thirty-four people, one pit, one class of injury.
• Jebel Sahaba: What a Restudy Did to the Oldest Massacre — the contested half of the record: how a 2021 re-examination of the same bones turned the oldest known massacre into a generation of repeated raiding, and why the certainty column exists.