> ## Content Index
> Fetch the complete content index at: https://www.meghantheengineer.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Harmony Gold Mine No. 4 Tailings Dam (South Africa, 1994)
- URL: https://www.meghantheengineer.com/publications/harmony-gold-mine-no-4-tailings-dam-south-africa-1994/
- Published: 2025-01-15T17:00:00.000Z
- Updated: 2026-09-05T20:41:13.000Z
- Description: Case study, with Riley Manwaring and Irfan Alvi. Association of State Dam Safety Officials, Dam Failures and Lessons Learned (damfailures.org), 2025.
- Author: Meghan Walter
- Tags: Publications, Failure case histories, Dam safety, Engineering ethics, #case-study, #Import 2026-09-05 12:49

On the evening of February 22, 1994, the northern wall of the Harmony Gold Mine No. 4 tailings dam breached above the suburb of Merriespruit, in the Free State goldfields of South Africa. A storm of ordinary size had overtopped a wall that had seeped and sloughed for years. About 630,000 cubic metres of slurry moved through the town in a wave 2.5 metres high. Seventeen people died, more than 200 were injured, and 80 homes were destroyed. The failure was fundamentally a human one, and it changed how South Africa regulates mine waste.

Key facts

| Item       | Detail                                                       |
| ---------- | ------------------------------------------------------------ |
| Location   | Merriespruit, Virginia, Free State, South Africa             |
| Type       | Upstream-method tailings dam, semi-dry paddock system, gold  |
| Height     | 31 m (102 ft)                                                |
| Built      | From 1978, about 300 m from the town                         |
| Failed     | February 22, 1994, about 9 p.m.                              |
| Mode       | Overtopping, then slope failure and static liquefaction flow |
| Released   | About 630,000 m³ through a 450 ft breach                     |
| Fatalities | 17, aged 2 to 71; more than 200 injured                      |
| Damage     | $17 million in 1994; about $35 million in 2023 dollars       |

## Background on tailings dams

Mine tailings, also called slimes, are the waste left after minerals are extracted and processed: a mixture of sand, silt, metals, chemicals, and water that may be solid, liquid, or a slurry of fine particles, and is often toxic and sometimes radioactive. Tailings are usually discharged as a slurry to a storage facility such as a tailings dam. A tailings dam begins as a starter dike sized to hold the first two or three years of tailings plus storm storage, and is raised as the tailings rise.

Dams are raised by one of three methods, named for the direction the crest moves relative to the starter dike. In the *upstream* method, tailings discharged around the crest form a beach, and the beach becomes the foundation for the next dike. In the *downstream* method, each raise is built on the downstream face. In the *centerline* method, fill is placed on the beach and the downstream slope at once. The upstream method is the least costly and the least stable; the downstream method is the most stable and the most expensive.

![](https://storage.ghost.io/c/5d/55/5d554874-dd60-43da-9b81-a65dd6989b32/content/images/2026/09/Figure-1---Construction-methods-selection--1-.png)

The three ways a tailings dam is raised. Redrawn from Troncoso, Verdugo, and Valenzuela (2017).

South African gold tailings dams are usually perimeter ring dikes raised by a variant of the upstream method called the daywall and nightpan paddock system. The outer perimeter is the daywall. It is raised to provide freeboard and to give the deposited tailings as long as possible, about three weeks on average, to dry and shrink in the sun. When a paddock has dried enough, the next paddock wall is built on it to receive the next batch. The rate of rise is typically one to three metres a year. This work is supervised during the day, hence the name. Tailings can also be discharged overnight, unsupervised, into the nightpan, from which they flow toward a penstock outlet tower, usually at the center of the dam, that decants water without disturbing the settled tailings.

![](https://storage.ghost.io/c/5d/55/5d554874-dd60-43da-9b81-a65dd6989b32/content/images/2026/09/Harmony7.png.webp)

Section through the daywall of a paddock-system tailings dam: stepped outer slope with a berm and drain every nine metres, low paddock walls holding fresh slurry, and the beach running back to the pool and penstock. Redrawn from Wagener (1997).

Control of the phreatic surface, the water table inside the dam, is the single most important factor in keeping a tailings dam stable and serviceable. Tailings are placed by open-ended pipe, by spigots along the wall, or by cyclones that separate the coarse and fine fractions. A dam may be operated as one or several compartments, each with its own delivery and decant system.

## History of the No. 4 dam

Merriespruit is a suburb of Virginia in the Free State goldfields, part of the Witwatersrand basin, the largest accumulation of gold in the world. Mining is South Africa's largest producer of solid waste. Virginia was laid out in 1936 after the goldfields were discovered, and the 250-home community of Merriespruit was built in 1956 to house mine staff. When work on the No. 4 dam began in 1978, about 300 metres from the houses, the town had already stood for two decades.

![Plan of the No. 4 dam and Merriespruit in 1985](https://damfailures.org/sites/default/files/2025-10/Harmony2.png)

Plan of the No. 4 dam in relation to Merriespruit, 1985\. The intended pool positions in compartments 4A and 4B are the hatched ovals. Source: Van Niekerk and Viljoen (2005).

The dam was designed by the mine's metallurgical manager with a representative of the tailings contractor. Tailings dam design in South Africa at the time was largely rule of thumb, and no geotechnical investigation was made. The dam had three compartments: 4A to the north, 4B to the south, and a small emergency compartment 4C farther south. There was no ring dike around the whole dam and no return-water dam; the penstock decanted into a sump that drained by gravity to the plant. Three pipelines delivered tailings. The No. 1 line gave trouble throughout the dam's life, with high friction head, frequent choking, and bursts, so the pool was hard to control from the start and often sat away from the penstock, which affected drainage.

Little is recorded of the day-to-day running of the dam, but the sequence of events before the failure is a record of warnings unnoticed or unheeded.

Warnings and decisions before the failure

| When            | What happened                                                                                                                                                                                                                                                                     |
| --------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| 1978            | Construction begins by the upstream paddock method, 300 m from Merriespruit; no geotechnical investigation                                                                                                                                                                        |
| 1985            | Consulting engineers assess the dam, then 17 m high at the future breach. Seepage found in a test pit at the critical section; factor of safety estimated at 1.4, falling to 1.2 at a projected 35 m. Judged in good condition                                                    |
| 1986            | Government notice exempts the mine from the Water Act dam safety regulations for five years                                                                                                                                                                                       |
| By 1991         | Sloughing on the northern wall worsens; a dumped rock buttress fails to stop it. Minerals Act requires storm storage with 0.5 m freeboard, but the exemption is extended to 1993                                                                                                  |
| Nov 1992        | Hydraulic buttress, 90 m long and 9 m wide, placed at the toe of the northern wall of 4A                                                                                                                                                                                          |
| 1993            | Four plants brought under one manager; fewer staff, less oversight, less training. Seepage above the drain exit patched. March: agreement to stop all deposition in 4A. Deposition continues anyway. Exemption extended to 1995\. April: operator's stability analysis gives 1.34 |
| Before Feb 1994 | Division wall between 4A and 4B breaches; 4B's drainage collects in 4A against the northern wall. Freeboard falls below 300 mm. Drain outlets clog                                                                                                                                |
| Feb 1, 1994     | Satellite image shows the pool about 0.45 m below the crest, ponded against the northern wall for months                                                                                                                                                                          |
| Feb 22, 1994    | 32 mm of rain, 20 mm of it in 20 minutes after 6 p.m. Water seen flowing over the northern crest at 7 p.m. Failure at 9 p.m.                                                                                                                                                      |

In 1985 consulting engineers assessed the dam. Eight piezometers were installed, three cone penetration tests run, test pits dug in the foundation, and samples tested for gradation, Atterberg limits, and shear strength. The site had about 1.5 metres of silty sand over clayey alluvium on weathered mudstone, siltstone, or sandstone. At the critical section, the site of the future breach, the pit met a drainage depression, considerable seepage, and 1.5 metres of loose clayey sand and very soft sandy clay over sandstone. The stability analysis gave a factor of safety of 1.4 at 17 metres and 1.2 at a projected 35 metres. The report concluded the dam was in good condition.

The next year a government notice exempted the mine for five years from regulations under the Water Act of 1956 that would have classed the dam as Category III on size and hazard and required maximum safety precautions. Why the mine was exempted, and then exempted twice more, has never been clear. Had it not been, the dam might not have failed.

By 1991 sloughing on the northern wall had increased, and a small rock buttress placed against the face did not stop it. The Minerals Act of that year required tailings dams to hold the 100-year, 24-hour storm with half a metre of freeboard; the exemption was extended two years. In November 1992 the operator built a hydraulically placed buttress at the toe of 4A's northern wall, 90 metres long and 9 wide, and deposited tailings there.

In 1993 the owner reorganized, placing four reduction plants under a single metallurgical manager who had not previously been responsible for Merriespruit. Staff at the dam were cut, managers oversaw more facilities with fewer hours each, and training suffered. Seepage appeared above the drain exit and was patched. In March the operator agreed to stop all deposition in 4A. Whether the order was unclear or ignored, tailings kept going in. The exemption was extended again, to 1995\. In April the operator's own analysis, using assumed strengths and a circular slip, gave a factor of safety of 1.34 for the northern wall.

Some time before the failure the division wall between 4A and 4B breached, so 4B's drainage flowed into 4A and collected against the northern wall instead of reaching the penstock. The pool grew, the freeboard at the northern wall fell below 300 millimetres, and over the months before the disaster the pool was pushed steadily toward that wall. Drain outlets were reported clogged by fungus, chemical precipitate, and mud. Investigators later identified the low relative density of the fill in the northern embankment, an inferior daywall built from slurry that in the early years often arrived thin, as a major cause.

## Failure and consequences

The summer of 1993 to 1994 was wet. On Tuesday, February 22, 1994, a storm dropped about 32 millimetres of rain, 20 of them in twenty minutes after six in the evening, an event with a five-year return period and, as Strydom and Williams[9](#ref-9) put it, nothing out of the ordinary. Witnesses saw water flowing over the crest of the northern wall at about seven. At nine the wall failed suddenly. A breach 450 feet wide released about 630,000 cubic metres of slurry through Merriespruit. The wave was 2.5 metres high when it reached the first houses and knocked some off their foundations. People already in bed found themselves floating against their ceilings; others were blinded by the mud.

![The No. 4 tailings dam after the failure](https://damfailures.org/sites/default/files/2025-10/Harmony4.jpg)

The No. 4 tailings dam after the failure, 1994\. Photo: Minerals Council South Africa.

Seventeen people died, aged 2 to 71\. More than 200 were injured, 80 properties were destroyed, and the environmental damage was wide. Losses came to $17 million in 1994 dollars, about $35 million today. Rescuers waded through chest-deep tailings containing cyanide and felt it burn their skin as they searched in the dark, with helicopters lighting the scene and ambulances shuttling survivors to hospitals.

![Aerial view of the breached embankment](https://damfailures.org/sites/default/files/2025-10/Harmony3.jpg)

The breached northern embankment from the air. Photo: Tailings.info.

![Ground view of the aftermath in Merriespruit](https://damfailures.org/sites/default/files/2025-10/Harmony5.png)

The aftermath in Merriespruit. Photo: Pretoria FM (2021).

The Minister of Justice convened a joint inquest and inquiry under a judge. Hearings opened on March 15, 1994, and judgment came the following year. The operator, the owner, and the state were represented, along with an insurer and the mineworkers' union. For the first time in South Africa, satellite imagery was entered as forensic evidence. In infrared images, wetter ground appears darker, and a frame from February 1, three weeks before the failure, showed the pool about 0.45 metres below the crest and ponded against the northern wall, where it had been for a long time. The law required freeboard for the 100-year storm; the dam did not have it. Operating staff denied the pool had sat against the wall until confronted with the imagery. With the eyewitness reports, it established overtopping as the critical trigger: erosion of the slope, progressive slope failures, and then static liquefaction flow of the metastable tailings.

On April 1, 1995, the judge found six employees of Harmony Gold Mine and of Fraser Alexander, the contractor that managed the dam, criminally liable. Three were convicted of culpable homicide, later reduced to contravening Section 37 of the Minerals Act. The fines totaled R270,000, under $15,000 at the time. The mine's profit in 1995 was about R56 million. The inquiry concluded that there had been sufficient legislation to prevent the disaster, if only it had been applied. South Africa drafted a mandatory Code of Practice on Mine Residue Deposits in 1995 and adopted it in 1998, with new standards for design, water management, qualifications of personnel, distance from people at risk, hazard classification of every tailings dam, and regular audits by experienced engineers.

The failure was fundamentally human. Had the dam been designed, built, maintained, or operated differently, it would not have failed; had it not been sited beside the town, or had the warning signs been heeded, no one need have died. Seepage and sloughing at the northern wall persisted for years despite repeated buttressing. Operations staff knew the freeboard in 4A was shrinking, but management apparently did not. Runoff had reached the town from the dam several times in earlier years, possibly from overtopping, without anyone recognizing that ponded water against the wall and erosion of the slope could bring it down. Through the same years the gold price fell and production costs rose, and the cost pressure that followed was a fundamental contributor to what happened.

## Human fallibility and limitations

People are fallible and limited, and that underlies every dam failure. It shows as misperception, faulty memory, vague language, incomplete information, lack of knowledge or expertise, unreliable intuition, inaccurate models, cognitive biases at the individual and group level, shortcuts, emotion, and fatigue. Here it appeared as errors across many people:

- Those who sited the dam beside the town either did not see the risk or were overconfident about managing it, and may not have understood the consequences of a major failure. The industry then believed gold tailings could not liquefy, which led to underestimating the chance of a flow that could reach the town.
- People building the dam knew the northern wall fill was loose, but the implications were not recognized.
- The geotechnical engineer overestimated the factor of safety of the northern wall, at 1.2 to 1.4, through lack of expertise or wrong assumptions.
- Those who tried repeatedly to stabilize the wall did not understand its physics well enough to succeed, and did not know the material could liquefy.
- Operations staff did not grasp the significance of continuing to place tailings-laden water in 4A against a stop-work order, which reduced freeboard and stability.
- Overtopping had not previously caused stability failures of tailings dams, so the industry did not recognize that it could.

## Risk management inadequacies

Human error and the pressures behind it lead to inadequate risk management, of three kinds. *Ignorance* is insufficient awareness of risk, from missing or wrong information, lack of expertise, unreliable intuition, or complexity. *Complacency* is awareness with too much tolerance, from fatigue, indifference, optimism bias, or pressure from other goals such as cost. *Overconfidence* is awareness with an inflated estimate of one's ability to cope.

At this dam the people involved were mostly ignorant of the risks because they lacked expertise. Their efforts to repair and buttress the northern wall show they saw a problem, but not its size. Continuing to pump into 4A after the stop-work order suggests operators did not understand pool management, freeboard, and overtopping, especially since overtopping had happened elsewhere, and probably here, without failure. Some may have understood the risk and accepted it under cost pressure, pressure from above to get rid of the water, or simple laziness, which is complacency. And siting a large tailings dam next to a town is overconfidence.

## Safety culture

Against those drivers stands safety culture: people at every level of an organization placing high value on safety, which produces a humble and vigilant attitude toward failure. It survives only when senior leadership visibly puts safety first and pays for it, in resources and in trade-offs. In hindsight, the safety culture of the South African gold industry, of Harmony Gold Mine, and of the people around this dam was lacking. As Wagener and others note, South Africa had been averaging two tailings dam failures a year, and Merriespruit still happened.

## Best practices

Strong safety cultures meet standard practice as a matter of course and often go beyond it. Failures are typically preceded by long neglect of many standard and best practices, and this one was no exception:

- **Design and construction.** The dam lacked a ring main and a return-water dam to capture runoff.
- **Design conservatism.** The northern daywall was too loose to have adequate shear strength, the history of erosion and sloughing suggests the slope was too steep, and the lack of freeboard let a storm of two inches overtop it.
- **Budget, resources, and resilience.** Cost pressure was fundamental. The dam was understaffed in management, engineering, and operations. Only one geotechnical engineer was involved, occasionally, at the contractor's head office; his role was vague and site staff did not keep him informed.
- **Humility, learning, and expertise.** A dam designed by a metallurgical manager, a wall no one could stabilize, and staff promoted beyond their training all point to a shortage of expertise.
- **Cognitive diversity.** One engineer, rarely present, is not diversity.
- **Decision-making authority.** Operations staff were able to keep pumping into 4A in violation of the stop-work order. Monitoring by management might have prevented it.
- **System modeling.** No one had thought through the failure modes, including the one that occurred: that overtopping could cause slope failure and that the tailings could liquefy and flow into the town. There was no emergency action plan, though two hours passed between overtopping and failure, enough to have evacuated the town had anyone been watching.
- **Information management.** Meetings between the contractor's local staff and the mine were poorly minuted and responsibilities were unclear, which produced grey areas and accusations of lying at the inquiry. The contractor's engineer was reprimanded for not climbing the dam to check the freeboard on a visit a few months before, but his own staff had not told him the dam was still receiving slime.
- **Warning signs.** They were seen, but their magnitude was not. Seepage, sloughing, remedial work, and a stop-work order show the risk of slope failure was recognized despite calculated factors of safety above 1.2\. What was not recognized was that the true factor was near 1.0, that ponded water had eaten the freeboard, that a moderate storm could overtop the wall, that overtopping erosion could trigger slope failure, and that slope failure could set off a liquefaction flow.
- **Standards.** The dam received repeated exemptions from regulation for reasons never explained, so the standard of care was not met. Without them, it is possible, perhaps likely, that a different course would have been taken.

The failure can ultimately be attributed to human factors: years of poorly judged managerial and operational decisions, rooted in a poor economic climate and cost-cutting. The operators lacked the expertise to address documented problems in design, construction, and operation, and never sought an unbiased outside assessment. They showed little understanding of the risks and consequences of failure, whether from inexperience or overconfidence. A breakdown in communication kept tailings flowing onto a structure known to be unsafe. An ordinary storm then did the rest: $17 million in damage and seventeen deaths, all of them preventable had the warning signs been recognized and heeded.

## Lessons learned

- [Dam incidents and failures can fundamentally be attributed to human factors.](https://damfailures.org/lessons-learned/dam-incidents-and-failures-can-fundamentally-be-attributed-to-human-factors?ref=meghantheengineer.com)
- [Intervention can stop or minimize consequences of a dam failure. Warning signs should not be ignored.](https://damfailures.org/lessons-learned/intervention-can-stop-or-minimize-consequences-of-a-dam-failure-warning-signs-should-not-be-ignored?ref=meghantheengineer.com)
- [Regular operation, maintenance, and inspection of dams is important to the early detection and prevention of dam failure.](https://damfailures.org/lessons-learned/regular-operation-maintenane-and-inspection-of-dams-is-important-to-the-early-detection-and-prevention-of-dam-failure?ref=meghantheengineer.com)
- [Static liquefaction should be considered as a potential failure mode for dams that have loose sands or silts in their embankments or foundations.](https://damfailures.org/lessons-learned/static-liquefaction-should-be-considered-as-a-potential-failure-mode-for-dams-that-have-loose-sands-or-silts-in-their-embankments-or-foundations?ref=meghantheengineer.com)
- [Timely warning and rapid public response are critical to saving lives during a dam emergency.](https://damfailures.org/lessons-learned/timely-warning-and-rapid-public-response-are-critical?ref=meghantheengineer.com)
- Tailings dams raised by the upstream method are more susceptible to liquefaction, especially when the tailings behind them are saturated.
- A tailings dam is under construction until it is closed, and needs continuing evaluation as conditions change.

## Further viewing

Tailings Dam Management for Engineers, Colorado School of Mines, Mining Engineering Department.

- [Intro to Tailings Dam and Coal Ash Impoundment Design, Construction and Monitoring](https://learningcenter.damsafety.org/products/intro-to-tailings-dam-and-coal-ash-impoundment-design-construction-and-monitoring-what-can-go-wrong-and-right-on-demand?ref=meghantheengineer.com), an ASDSO on-demand webinar.

## References

1. Piciullo, L., Storrøsten, E.B., Liu, Z., Nadim, F., and Lacasse, S. (2022). A new look at the statistics of tailings dam failures. *Engineering Geology*, 303.
2. Troncoso, J.H., Verdugo, R., and Valenzuela, L. (2017). Seismic performance of tailings sand dams in Chile. *Proceedings of the 16th World Conference on Earthquake Engineering.*
3. Vermeulen, N.J. (2007). *The composition and state of gold tailings.* PhD thesis, University of Pretoria.
4. Wagener, F. (1997). The Merriespruit slimes dam failure: overview and lessons learnt. *Journal of the South African Institution of Civil Engineering*, 39(3).
5. Williams, D.J. (2021). Lessons from tailings dam failures: where to go from here? *Minerals*, 11(8), 853.
6. Van Niekerk, H.J., and Viljoen, M.J. (2005). Causes and consequences of the Merriespruit and other tailings dam failures. *Land Degradation and Development*, 16(2), 201 to 212.
7. Hansen, R.N. (2015). Contaminant leaching from gold mining tailings dams in the Witwatersrand Basin, South Africa: a new geochemical modelling approach. *Applied Geochemistry*, 61, 217 to 223.
8. Wagener, F.V.M., Craig, H.J., Blight, G., and McPhail, G. (1998). The Merriespruit tailings dam failure: a review. *Tailings and Mine Waste '98*, 925 to 952.
9. Strydom, J.H., and Williams, A.A.B. (1999). A review of important and interesting technical findings regarding the tailings dam failure at Merriespruit. *Journal of the South African Institution of Civil Engineering*, 41(4), 1 to 9.
10. Gold FM 104.3 Welkom. (2021, February 22). Today we remember the Merriespruit disaster. Facebook.
11. Tailings.info. (n.d.). Merriespruit tailings dam failure, Virginia, South Africa. Tailpro Consulting.
12. Minerals Council South Africa. (n.d.). We care and we remember: Merriespruit, 22 February 1994.
13. Pretoria FM. (2021, February 22). Photograph. Facebook.
14. Blight, G., and Fourie, A. (2014). *A review of catastrophic flow failures of deposits of mine waste and municipal refuse.* University of the Witwatersrand.
15. Staff reporter. (1996, March 22). White-collar killers walk free. *Mail and Guardian.*
16. *Martin Creamer's Mining Weekly.* (1997). 14(23), 13.
17. Alvi, I. (2013). Human factors in dam failures. *ASDSO Annual Conference Proceedings.*
18. Alvi, I. (2024). Dam incidents and failures can fundamentally be attributed to human factors. Association of State Dam Safety Officials.
19. Alvi, I.A., and Alvi, I.S. (2023). Why dams fail: a systems perspective and case study. *Civil Engineering and Environmental Systems*, 40(3), 150 to 175.

This case study was written for the Association of State Dam Safety Officials with Riley Manwaring, Gannett Fleming, and Irfan Alvi, Alvi Associates, and peer reviewed by Dino Bernardi, P.E., California Division of Safety of Dams.

[Read the case study at damfailures.org](https://damfailures.org/case-study/harmony-gold-mine-no-4-tailings-dam-south-africa-1994?ref=meghantheengineer.com)