After Goliath:
how rare are fatal roller-coaster ejections?

Families around Atlanta are asking a reasonable question: how can someone leave a roller-coaster seat, and how often has it happened before? Behind every case is a person and a family. We want to answer carefully, without turning their loss into spectacle or filling gaps with guesses.

This page keeps those questions separate. You can inspect the source for every listed case, filter the inventory, and download the records. The inventory is a starting point for research, not a claim that every American case has been found.

Six Flags Over Georgia · Developing investigation

What has been established about Goliath?

1. A distinction in early injury reporting

   October 12, 2026 at 7:28 PM Eastern

   WSB-TV reports four other people injured. AP’s police-attributed report uses the broader wording ‘several.’ We are preserving that distinction rather than presenting an early count as a final medical tally. Neither report establishes why the incident happened. The historical case inventory remains separate from this developing investigation.

   WSB-TV — one death, four injuries reported · Associated Press — police-attributed report

2. Initial evidence review: cause not established

   October 12, 2026 at 7:10 PM Eastern

   Cobb County police confirmed one death and several injuries after an incident at Six Flags Over Georgia on October 12, according to AP and local reporting. FOX 5 reported witness accounts of a rider falling from Goliath. Those accounts do not establish why the person fell. The sources reviewed did not contain a final official cause determination. Goliath is not included in the historical inventory below.

   Associated Press — October 12 report · FOX 5 Atlanta — developing Goliath coverage

Questions a completed investigation may answer

• What sequence of events led to the fatality and other injuries?
• What do physical evidence, restraint checks, ride records and witness interviews establish?
• Were any corrective measures ordered, and what findings support them?

A reported fall is not, by itself, proof that a restraint opened, a part failed, an operator erred or a rider disregarded instructions. An investigation can also end without a definitive explanation. We will identify that outcome honestly if it happens.

The National Safety Council’s survey for IAAPA estimates the following for 2024, United States and Canada, permanent amusement facilities. These are not U.S.-only coaster-ejection figures.

The overall injury rate was 0.78 per million rides. Coaster injuries—not fatal ejections—were estimated at 1.27 per million coaster rides. The report supplies no separate fatal-ejection rate.

This is a weighted survey, not a complete incident census: 227 facilities responded, 164 supplied ridership, and reported rides covered about 65% of the estimated total. It excludes traveling carnivals and water parks; miniature coasters can fall under children’s rides.

Read the NSC/IAAPA 2024 public report (PDF). These figures provide broad injury context, not a finding about Goliath’s safety.

A historical study asked a closer question

Pelletier and Gilchrist’s peer-reviewed study covered U.S. roller-coaster-associated deaths from May 15, 1994, through May 14, 2004. It identified 40 deaths: 29 patrons and 11 employees. Nine patrons fell or were ejected; 18 patron deaths involved medical conditions and two involved other external injuries.

The researchers combined federal records, news, medical literature and regulator information. They warned that cases could be missed, that source classifications might include unconventional rides, and that temporal association does not establish medical causation. They could not calculate fatality risk without adequate ridership estimates. Read the full historical study (PDF).

Those nine cases belong to that study’s period and definitions. We do not add its total to our inventory, use it as today’s annual rate, or treat all coaster-associated deaths as ejections.

Why we will not manufacture “one in X”

To calculate a meaningful probability, both sides must match: verified fatal ejections during a defined period, divided by coaster rides during that same period and geography, using consistent definitions. Dividing decades of selected news cases by one recent year of all amusement rides would produce a precise-looking but invalid answer.

“Rare” also does not mean unimportant. A small population-level risk cannot explain an individual loss, dismiss a family’s questions, or settle a ride’s condition.

Our working definition is a fatal departure from a roller-coaster seat, including reported falls and ejections. Collision ejections and departures followed by strikes or entrapment are labeled separately. This broader definition avoids quietly treating unlike events as identical. Employees are a separate filter.

Not exhaustive. The selected records span 1964–2013; that date range is not a claim of complete coverage. Goliath’s October 2026 incident remains in the developing log above. One row represents one incident; Wild Wonder involved two deaths.

How many involved Six Flags?

Within this inventory, four guest deaths in four pre-2026 incidents occurred at parks operated by Six Flags at the time: Rail Blazer (1984), Lightnin’ Loops (1987), Superman: Ride of Steel (2004) and Texas Giant (2013). The employee test-ride fall on Rolling Thunder in 1981 is separate. Use the operator and people filters to inspect the supporting rows.

This is a count of the documented selection, not a certified lifetime corporate total or a comparison of operators’ safety. Companies buy, sell and manage parks; today’s branding can be a misleading guide to who operated a park decades earlier.

For example, Colossus’s 1978 death preceded Magic Mountain’s 1979 acquisition by Six Flags. Darien Lake was managed by Herschend for CNL in 2011. We do not assign those incidents to Six Flags because of later branding.

What past investigations can—and cannot—teach

The Massachusetts report on the 2004 Superman fatality examined restraint fit, admission and checking. A mechanism can function mechanically while failing to hold a particular rider adequately. That is a finding about that case, not evidence about what happened on Goliath.

In contrast, the Louisiana State Fire Marshal’s final Xtreme release said investigators could not determine the exact cause of the 2010 fall. Precautionary changes were still required. A final report can document useful findings without explaining every part of an event.

We will not use these cases to infer a shared cause, rank manufacturers, or blame someone in the current investigation. Lawsuit allegations, witness recollections, a park’s statement and an investigator’s final conclusion are different kinds of evidence.

Our research and update protocol

1. Start with the question and scope. U.S. fatal rider departures from roller coasters. Keep employees, collision ejections and mixed departure/strike events identifiable. Exclude medical deaths without a rider departure, unrelated ride types, and people struck while entering a ride area.
2. Follow the evidence back. Prefer official investigative records. Use contemporaneous local or wire reporting where records are unavailable. Link the source, retain its date, and distinguish reporting from a final finding. Syndications of one story are not independent confirmation.
3. Record uncertainty. Missing cases, unresolved dates, uncertain mechanisms and changing ownership prevent a complete national total. Date precision follows the source. We are still seeking stronger records for older leads, including Lake Compounce’s Wildcat and Magic Harbor’s Black Witch; they are not counted here.
4. Keep the current incident separate. Add material, sourced developments to the dated Goliath log. State what changed and what remains unknown. Do not turn silence, a reopening or a settlement into a cause determination.
5. Correct visibly. Publish corrections as new dated entries identifying the earlier claim. Preserve the history, rather than silently changing an explanation. Historical inventory revisions need their own version and scope review.
6. Treat people with care. No graphic imagery, speculative personal details or inferred fault. A person’s size, disability or behavior is not a substitute for evidence. Read official findings in their full context.

Why there is no single easy federal total

The CPSC distinguishes permanent amusement rides from mobile rides in its jurisdiction guidance. Fixed-site oversight involves state and local arrangements. That does not mean permanent rides are unregulated; it helps explain why records must be assembled from multiple sources. General emergency-department amusement-ride estimates are not a fatal-coaster-ejection census.

This research combines a selected historical inventory with published studies. It is not a systematic review of every newspaper archive or every state record, and no absence of a row establishes that an event never happened. Reader leads can expand it, but they must be checked before counting.

Found a missing case or a stronger source? Send a correction with the ride, date and original record. Please leave out private family information. Submissions do not publish automatically.

Start your own research here

• NSC/IAAPA, 2024 North America Fixed-Site Amusement Ride Safety Report — survey scope, estimates and injury definitions; this is 2024 data, regardless of its upload date.
• Pelletier & Gilchrist, Injury Prevention (2005), DOI 10.1136/ip.2005.008425 — historical fatality study; full-text reprint.
• Massachusetts Department of Public Safety, Superman investigation — primary report hosted in an accident-record archive.
• Louisiana State Fire Marshal, November 19, 2010 final release — an official inconclusive investigation.
• The developing-news links in the update log and the source beside each inventory row provide the supporting record for those particular claims.

Editorial note: This free research feature is published by a local family business. We are not ride engineers, investigators or a regulator. AI assisted with research organization and drafting. The linked evidence—not our business identity—is the basis for evaluating the claims.