The Bermuda Triangle is famous for swallowing ships and planes. I put 472 famous disappearances on the map and tested every sea against its shipping. The Triangle does have more than its traffic predicts, about 2.9 times, but so does the whole American side of the Atlantic. The places that really are worse than their traffic are less famous, and easier to explain.
Draw a line from Miami to Bermuda, on to San Juan in Puerto Rico and back to Miami, and you have the most famous patch of sea in the world. Since the 1960s, books and films have said that ships and aircraft vanish there more often than anywhere else. The usual reply is that the Triangle is simply busy: lots of traffic, so lots of losses. That is a testable claim, so I tested it.
I took Wikipedia's lists of missing ships, missing aircraft and people who disappeared mysteriously at sea, added the incidents told in the articles on the Bermuda, Dragon's and Michigan Triangles, merged the duplicates and placed each disappearance as well as its text allows: 673 incidents, 472 of them at sea. Then I compared each sea with its shipping, measured by the World Bank and IMF's map of ship positions from 2015 to 2021, and with all 17,224 recorded wrecks that have a position. If a sea really is cursed, it should hold more disappearances than either would predict.
The three lists are a mixed bag: 203 ships and boats, from Tudor warships to yachts; 290 aircraft, from Victorian balloons to Malaysia Airlines Flight 370; and 180 people, from explorers to cruise passengers, who vanished over the side or with their boat. 61 rows turned out to be the same incident listed twice. The median year of a placed incident is 1940, and 69% are from before 1946, so most of these losses happened long before modern shipping lanes existed; 25% happened during the two world wars.
Placing them is the hard part. Only 37 give coordinates or a distance from a named place. 220 say "off" or "near" a cape, island or port; 52 name only the two ports between which a ship vanished, so I use the midpoint of the route; and 163 name only a sea. I left out the 201 that say no more than "Atlantic Ocean", or that were lost over land.
Famous disappearances per decade, 1800 to 2025
The 472 incidents placed at sea, by what was lost. Hover a bar.
Each dot is a disappearance, coloured by what was lost or by how precisely I could place it. The blue haze is modern shipping. Each sea is shaded by how many disappearances it holds compared with the number its share of shipping predicts: red for more, blue for fewer, pale where the difference is within chance. Switch the shading to the other two tests, or to 2 degree squares, and click anything.
Dots: 472 disappearances placed at sea (Wikipedia lists and the three triangle articles, duplicates merged). Shading: observed disappearances (or losses) divided by the number expected from the sea's share of modern AIS traffic (or of all recorded losses since 1800); pale where the 95% interval includes 1. Squares use only incidents placed by a position, a named place or a route. Traffic: World Bank, Global Shipping Traffic Density (IMF and World Bank, AIS positions 2015 to 2021), CC BY 4.0, shown as a modern proxy only. Recorded losses: Wikipedia lists of shipwrecks and Wikidata. Outlines: Natural Earth. Base map tiles © Esri.
The Triangle covers about 1,136,000 square kilometres and carries 0.59% of the world's tracked shipping. If disappearances simply followed traffic, it would hold 2.8 of the 472 placed incidents. It holds 8: 2.9 times as many, with a 95% interval of 1.2 to 5.6 times. That is more than chance would usually give, but only just, and it is not unusual for its part of the world. The four seas around it, the US Atlantic coast, the Caribbean, the Gulf of Mexico and the Grand Banks, hold 80 disappearances against 29.1 expected, 2.7 times their traffic. Within that group the Triangle has 8.8% of the shipping and should hold 7.7 of their disappearances. It holds 8.
Slightly, but not because it is the Triangle. Its excess, about 2.9 times, matches the excess of the whole western North Atlantic, which is what you would expect from lists written mostly in English, about American and British ships and flights, with a long tail of light aircraft out of Florida. Compared with its neighbours it is exactly as dangerous as its shipping suggests. Its few modern losses (1 recorded since 2000, against 2.6 expected) are no worse than anywhere else.
The legend also leaks at the edges. Of the 14 incidents told on Wikipedia's Bermuda Triangle page, only 2 lie inside the triangle as usually drawn, and both of those are placed only at the midpoint of their route. Flight 19 vanished off Florida's east coast, just north of the Miami to Bermuda line; the Star Tiger was lost on the way from the Azores, far out in the Atlantic; the Cotopaxi's wreck lies off St Augustine; the Marine Sulphur Queen went down in the Gulf of Mexico; the Carroll A. Deering ran aground at Cape Hatteras. 8 of the 14 are within 300 km of it. The Triangle has grown in the telling to cover most of the western North Atlantic.
The incidents on the Bermuda Triangle page
Where I placed them, and how far outside the Miami, Bermuda and San Juan triangle they fall. Click a row to see it on the map.
There is one test the Triangle does fail. Measured against all recorded wrecks with a position since 1800, it has 6.5 times more disappearances than expected (8 against 1.2). But that says more about the wrecks than the mysteries: only 44 recorded wrecks there have a position, because ships lost in deep open water leave no wreck to plot, and wartime sinkings, which dominate the positioned wrecks, were rarer there than off the American coast. A ship that sinks over the Puerto Rico Trench simply vanishes. That is geography, not mystery.
Here is every sea in the comparison, from the busiest (the Mediterranean and the North Sea) to the ones where few ships go now. Each bar is a 95% interval for the ratio of disappearances to the number expected; anything whose bar crosses 1 is within chance. Switch between the tests. The Mediterranean, the North Sea, the Baltic and the seas around Japan and China have far fewer famous disappearances than their shipping predicts. The open North Atlantic, the South China Sea and the Gulf of Mexico are about as expected. The Bermuda Triangle sits among its neighbours, well below the seas that really stand out.
Disappearances per unit of traffic, by sea
Ratio of observed to expected, log scale, with exact Poisson 95% intervals. Hover a row.
A funnel plot shows the same thing with the role of chance made explicit. Seas where little is expected scatter widely by luck alone; seas where a lot is expected should sit close to 1. The funnel lines mark where 95% and 99.8% of seas would fall if disappearances simply followed traffic.
Funnel plot: disappearances against traffic
Each point is a sea (or a 2 degree square): expected count across, observed over expected up, both on log scales. Hover a point.
7 seas hold clearly more disappearances than their traffic predicts, beyond the 99.8% funnel. None of them is a legend, and each has a plain explanation. Bass Strait and the Tasman Sea top the list, with 48 incidents against 0.18 expected: colonial schooners and packets lost between Sydney, Hobart and Port Phillip in the 1800s, in one of the stormiest straits in the world, carefully gathered in one section of the list of missing ships. Even with a traffic measure that is kinder to the Southern Hemisphere it is 10.6 times. Lake Michigan and the other Great Lakes come next: lake freighters lost in autumn gales a century ago, when lake trade was far busier than it is now.
Seas clearly worse than their traffic predicts
Beyond the 99.8% funnel limit. Click a row to zoom the map there.
At the scale of 2 degree squares, 28 squares are beyond the 99.8% limit with at least two incidents. 11 of them are in Bass Strait and the Tasman Sea. Most of the others are about aviation: the approaches to Hawaii, the Dole Air Race of 1927, when three of the planes racing from California to Hawaii vanished, and pioneer flights from Newfoundland. Aircraft follow routes that ships rarely use, so measuring them against shipping makes them look worse than they were. Just 1 of the squares is inside the Bermuda Triangle: the Bahamas, where a Flying Boxcar and a Cessna vanished, among thousands of light aircraft that cross from Florida every year.
The 2 degree squares furthest beyond their traffic
At least two incidents and beyond the 99.8% limit, most extreme first. Click a row to zoom the map there.
The seas that really are worse than their traffic are not mysterious: they are stormy, shallow, busy in the past, or very well written about.
The Dragon's Triangle, or Devil's Sea, south of Tokyo, has no agreed shape; Wikipedia's article says its outline "varies from one author to another". I drew it from Tokyo to Guam and to Taipei, a generous version around the article's centre point. It holds 7 placed disappearances against 6.3 expected from its traffic (1.1 times, 95% interval 0.4 to 2.3): exactly as many as its busy shipping predicts. 4 of them are from the Second World War, among them the giant carrier Shinano, sunk by a submarine on her first voyage. Its most famous case, the research ship Kaiyo Maru No. 5, was destroyed in 1952 by the undersea volcano she had gone to study.
The Michigan Triangle runs from Ludington to Manitowoc and down to Benton Harbor. By my placing, none of the 6 incidents in its Wikipedia article falls inside it, though 4 are within 50 km. The incidents placed only as "Lake Michigan" do sit at a point inside it, so I do not count them in the triangle. Lake Michigan as a whole, though, is genuinely worse than its traffic: 16 disappearances against 0.23 expected, 11 of them before 1946. Today's lake traffic is a shadow of the schooners and steamers of the 1800s, and the lake's autumn storms were notorious. Against all recorded losses, Lake Michigan's excess shrinks to 7.5 times.
Comparing old disappearances with modern traffic mixes eras. The fairest test compares like with like: the 433 wrecks from 2000 to 2025 that have a position in Wikipedia's lists or Wikidata, against shipping from 2015 to 2021. Here the Bermuda Triangle has 1 wreck against 2.6 expected. The seas that stand out are the Gulf of Alaska (34.9 times) and the US Atlantic coast (14.7 times), and the reason is again who writes it down: 100% of the Alaskan losses are fishing boats entered, with positions, into Wikipedia's monthly shipwreck lists, and 55% of all positioned losses since 2000 are in North American waters. Switch the chart above to "Losses 2000 to 2025 v traffic" to see every sea.
Disappearances. I parsed the tables of three English Wikipedia lists (List of missing ships, List of missing aircraft, List of people who disappeared mysteriously at sea; 201, 289 and 223 rows, leaving out the people list's solved cases) and added 21 incidents told in the articles on the Bermuda Triangle, the Devil's Sea and the Lake Michigan Triangle. Rows were merged when they shared a year and a distinctive name word, or an aircraft and a person shared a full date and a place, giving 673 incidents. Each was placed by a hand-made gazetteer of ports, capes, islands and seas: coordinates or a distance and bearing where given ("position"), "off" or "near" a named place, the great-circle midpoint of a route of up to 4,000 km, or a representative point in a named sea. Longer routes count only for the open sea they cross. Points that fell on land were moved to the nearest sea within 75 km (200 km for routes); 42 were moved.
Traffic. The World Bank's Global Shipping Traffic Density counts AIS positions from all vessels, January 2015 to February 2021. I used its 0.08 degree overview, which samples single pixels, so I only use it summed over 2 degree squares or whole seas, as a share of the world total. AIS positions favour coasts with dense receivers, ports and slow ships, so I repeat every test with a "lanes" index: the share of sampled pixels with at least 1,000 positions, which is less sensitive to how many positions are picked up. It is a modern proxy only: it cannot tell us where ships sailed in 1850.
Recorded losses. As a historical denominator I use the 17,224 wrecks with a position from Wikipedia's lists of shipwrecks (15,133) and Wikidata (2,091), 15,761 of them dated 1800 to 2025. I leave out the national bulk records (Scotland's Canmore, Wales's Coflein, the United States' AWOIS), which would flatter the seas of three countries. 69% of these positioned wrecks are war losses, so I also repeat the test without the war years.
Statistics. For each sea, the expected count is the total multiplied by the sea's share of traffic (or of recorded losses in the same years). Intervals are exact Poisson 95% intervals; "clearly" more or fewer means beyond the 99.8% funnel limit. The seas are simple outlines that I drew; each point belongs to the first sea that contains it, triangles first. Incidents placed only by a sea's name are never counted inside a triangle.
What this cannot say. These are famous disappearances: the ones somebody wrote about, in English, on Wikipedia. They lean towards the English-speaking world, aviation and the twentieth century. Many placements are rough, and a different outline for a triangle would change its count. Disappearances also cluster (one storm, one air race), which makes the intervals a little too narrow. The honest conclusion is about the pattern, not the second decimal place.