AnomalyRegistry
AR-0017

Marfa Lights

Partially explained
Date Reported since the late 19th century. Principal study 2004.
Location Mitchell Flat, east of Marfa, Presidio County, West Texas
Coordinates 30.3092, -103.9344

Summary

Stand at the viewing area on Highway 90 outside Marfa, look southeast across the desert, and on many nights you will see lights. They hover. They split in two. They merge. They drift, brighten, and go out.

In 2004 a group of physics students from Texas State University did the obvious thing, which nobody had done in a hundred and twenty years. They put cameras at the viewing area, and at the same time they watched the traffic on Highway 67, which runs across the flat about thirty kilometres away, on the far side of the desert.

Every time a car appeared on the highway, a Marfa light appeared. Every time a car went behind a rise, a Marfa light went out.

what nobody had done in 120 years the viewing area, Hwy 90 ~30 km of desert. layered air. Hwy 67 a car A MARFA LIGHT a car appears on 67, a light appears. the car goes behind a rise, the light goes out. repeatable. Texas State, 2004. the hovering, the splitting and the merging are not properties of the light. they are properties of the air between you and it.
Figure Drawn by Anomaly Registry from the 2004 Texas State University observation. Cameras at the viewing area and simultaneous observation of traffic on US Highway 67, about 30 km away across Mitchell Flat. A car appears, a light appears; the car passes behind terrain, the light goes out.

What is documented

The lights. Seen from the official viewing area on Highway 90, east of Marfa, looking southeast across Mitchell Flat toward the Chinati Mountains. They appear as glowing points, usually white or yellow, sometimes red. They appear to hover, to move laterally, to divide, and to combine. They are seen on many nights of the year and they are seen by a great many people.

The tradition. The lights are conventionally said to have been first reported in 1883 by a young cowhand named Robert Reed Ellison. The registry notes that this attribution is not well documented, that the account is a later one, and that origin stories of this kind are frequently retrofitted to establish that a phenomenon predates the obvious modern explanation for it.

The 2004 study. The Society of Physics Students at Texas State University, working with Karl Stephan, conducted a systematic observation.

They placed instruments at the viewing area. They simultaneously observed United States Highway 67, which runs from Marfa southwest toward Presidio, crossing the field of view at a distance of roughly thirty kilometres, on the far side of Mitchell Flat.

The correlation was direct and repeatable. A vehicle appears on Highway 67, and a light appears in the position where a Marfa light would appear. The vehicle passes behind terrain, and the light goes out.

They demonstrated it, on multiple occasions, and they published it.

Why the lights behave so strangely. The line of sight from the viewing area to Highway 67 passes low over a large flat expanse of desert.

Desert air on such a plain forms strong thermal layers, particularly at night when the ground cools rapidly and the layers invert. Light passing horizontally through those layers is refracted, and the refraction is not steady: it shifts, it magnifies, it splits an image into two, it makes a point of light appear to hover well above where it is, and it makes it wobble and swim.

A pair of headlights thirty kilometres away, seen through a shimmering wedge of desert air, does not look like a pair of headlights. It looks like a light that hovers, splits and merges.

Leading explanations

Vehicle headlights on Highway 67, refracted over Mitchell Flat. Established for the bulk of the sightings.

The 2004 observation is not an inference. It is a simultaneous measurement of the cause and the effect, and it is repeatable by anybody willing to sit in the dark with a camera and a view of the highway.

The residue. James Bunnell, an engineer who has monitored the flat for years with instrumented stations, has argued that a small number of the lights he has recorded do not correlate with traffic and have properties that headlights do not.

The registry records this claim and does not endorse it. It has not been independently confirmed, and it is the sort of claim that requires a great deal of confirming, because the null hypothesis, that everything is headlights, has been demonstrated for the great majority of events.

Anything predating cars. If lights were genuinely seen on Mitchell Flat before there were headlights, that would be significant. The evidence for it is a nineteenth-century attribution that first appears in the twentieth century, and the registry treats it accordingly.

What the popular version gets wrong

"They cannot be car headlights, because the lights were seen before cars existed." The claim that the lights were seen before cars existed rests on a single origin story of poor provenance. The claim that they are car headlights rests on somebody watching the cars and the lights at the same time.

"They move in ways headlights cannot." They move in ways headlights appear to move when their light travels thirty kilometres horizontally through layered desert air at night. The hovering, the splitting and the merging are not properties of the light source. They are properties of the atmosphere between you and it.

"Scientists have never been able to explain them." A group of undergraduates explained them in 2004, using cameras and a road map, and the reason it took a hundred and twenty years is that nobody had thought to look at the highway.

Current status

Partially explained. The mechanism for the great majority of sightings is established by direct simultaneous observation. A residue is claimed and is not established.

The registry keeps this record beside AR-0087 and AR-0032, because they are the same story in different clothes. The Devil's Kettle lasted eighty years because nobody put a flow meter in the river. The sailing stones lasted sixty because nobody put a camera on the playa.

The Marfa lights lasted more than a century because nobody turned around and looked at the road.

Sources

  • Society of Physics Students, Texas State University, with Stephan, K. (2004). Simultaneous observation of the Marfa lights and of traffic on US Highway 67.
  • Bunnell, J. Instrumented monitoring of Mitchell Flat and published claims of non-traffic events.
  • Literature on atmospheric refraction, thermal inversion layers, and superior mirage over flat terrain.

Last reviewed: July 2026. Records are provisional. Where the evidence changes, the entry changes. Found an error? Tell us.

Advertise on Anomaly Registry

A documentary archive of 100 investigated cases, read slowly and at length by an audience that arrives on purpose. Below are the current delivery figures and what is available to buy.

Figures are rolling daily averages, current as of September 2026. They are restated as the traffic changes rather than left to age quietly, and we are happy to screen-share the underlying Ad Manager and analytics reporting before an insertion order is signed.

Anomaly Registry is a reference archive rather than a feed. Every one of its hundred records is written to a fixed structure, sourced to contemporaneous documents, inquest files, peer-reviewed literature or credible reporting, and each carries a section setting out precisely what the popular version of the story gets wrong. Cases that were later explained and cases that were comprehensively debunked are kept alongside the unsolved ones, because an archive that keeps only its mysteries is not an archive. That editorial position is the reason people stay.

It also explains the shape of the traffic. A session here runs to more than four pages and close to nine minutes, which is several times what a typical content site sees. Readers arrive looking for one case, find the sources laid out, and follow the cross-references into the next record. For an advertiser, the practical consequence is that a banner on this site is in front of a settled, attentive reader for minutes rather than seconds, and it is seen repeatedly across a single visit without ever being the reason someone leaves.

Three placements are available on every page. The leaderboard sits above the masthead and accepts 728x90, 970x90 and 320x50. The in-article unit sits inside the body of the record, where the reader is already stopped, and accepts 728x90, 300x250 and 336x280. The footer unit closes the page and accepts 728x90, 970x90 and 320x50. There are three slots per page and there will not be a fourth: the inventory is deliberately limited so that each impression is worth more than it would be on a page carrying eight.

What we do not run matters as much as what we do. There are no interstitials, no pop-ups, no auto-playing video, no content-recommendation chumboxes, and nothing that reflows the page after the reader has started reading. Ads are served in reserved space so the layout does not jump. This protects viewability for the advertiser as directly as it protects the reading experience, and it is the reason we can quote the session figures above with a straight face.

The editorial environment is unusually safe for a subject area that is often anything but. This site does not publish conspiracy content, it does not promote pseudoscience, and it does not treat a rumour as a finding. Its entire method is to separate what is documented from what has merely been proposed, and to say plainly when a famous mystery has an unglamorous answer. A brand placed here sits next to sourced, corrected, carefully hedged writing rather than next to speculation dressed as news.

The readership is a general-interest one with a research habit: people who read long-form nonfiction, follow science and history coverage, and check citations. In practice the categories that fit well are books and publishing, documentary and streaming, museums and exhibitions, education and online courses, audio and podcasts, consumer technology, outdoor and travel, and financial services. If your product rewards attention and explanation rather than impulse, this is a good room to be in.

Everything is delivered through Google Ad Manager, so you can buy the way you already buy. Direct-sold campaigns can be trafficked as a guaranteed line item with standard reporting on impressions, viewability and click-through. Programmatic demand can reach the same inventory through the open auction or through a preferred deal or private marketplace at a floor we agree in advance. We accept standard IAB display creative and HTML5, and we can accommodate a reasonable third-party ad server or verification tag.

Category sponsorship is available for advertisers who want something more durable than a rotation. The register is organised into seven categories, and a sponsor can take the whole of one, appearing across every record inside it with an acknowledgement in the page furniture. Sponsorship never buys editorial influence: a sponsor cannot commission a record, cannot review one before publication, cannot have a record altered, and cannot have one removed. That rule is not negotiable and it is the thing that keeps the inventory worth buying.

Some categories are declined outright, and it is fairer to say so up front than to discover it after an insertion order. We do not accept adult content, gambling, payday or high-cost short-term credit, cryptocurrency speculation and token sales, dietary supplements making health claims, political advocacy, psychic or clairvoyant services, or anything that reads as a scare advertisement dressed up as an article. Nothing may be styled to look like a record or like part of the editorial page.

Rates depend on placement, volume, whether the buy is run-of-site or targeted to a category, and how long it runs. There is no rate card on this page because a sensible number depends on what you are trying to do, and quoting one in the abstract usually wastes both parties' time. Tell us the campaign, the flight dates and the budget you are working to, and you will get a straight answer including whether we think the fit is wrong.

To start, write to or use the contact form and mark the message for the advertising desk. Include the flight dates, the placements you want, the creative sizes you have ready and any verification requirements, and we will come back with availability and a quote. Media buyers who want the underlying analytics before committing are welcome to ask; we would rather show the reporting than be taken on trust.

Privacy Policy

This policy explains what information Anomaly Registry collects when you visit this website, why it is collected, how it is used, and what choices you have. It applies to anomalyregistry.com and to any subdomain operated by us. It was last updated on the date shown at the end of this policy, and we will revise it if our practices change.

We have designed this site to collect as little personal information as possible. You do not need an account to read anything here. There is no login, no user profile, and no comment system. In the ordinary course of reading the register, you are not asked to identify yourself, and we do not ask you to.

Like virtually every web server, ours produces access logs. These logs record the internet protocol address of the requesting device, the date and time of the request, the page or file requested, the referring page if one was sent, and the browser user agent string. These logs exist for security, abuse prevention, and diagnosing technical faults. They are retained for a limited period and then discarded. We do not use them to build profiles of individual readers.

If you write to us through the contact form, we receive whatever you choose to put into it: typically a name, an email address, and a message. We use that information for the sole purpose of reading and answering your message. Writing to us does not add you to any list, and we do not sell, rent, or trade contact details with anyone.

There is one optional mailing list, and you only join it by asking. If you submit the subscribe form we store the address you gave, the time you submitted it, the IP address it came from, and the exact wording of the consent you agreed to. We keep that record because Canadian anti-spam law expects us to be able to show when and how you asked to hear from us. The list is used to say that records have been published or materially revised, and for nothing else. Every message carries a way to leave, and asking to be removed by email works just as well.

Corrections are a special case, and worth stating plainly. If you write to point out an error in a record, we may act on that correction and revise the entry. We will not publish your name or your email address in connection with a correction unless you explicitly ask us to be credited.

This site is supported by advertising served through Google Ad Manager. Advertising technology providers may set and read cookies or similar identifiers on your device in order to select advertisements, limit how often you see the same advertisement, and measure performance. These identifiers are set by those third parties, not by us, and the data they collect is governed by their own privacy policies rather than this one.

We do not have access to the personal data that advertising partners collect, and we cannot delete it on your behalf. If you wish to limit interest-based advertising, you can adjust the advertising settings offered by Google, use your browser's cookie controls, or use the industry opt-out tools that are available in your region. Details are set out in our Cookie Policy.

We may use privacy-respecting aggregate analytics to understand which records are being read and how people move through the register. Where we do, we configure it to avoid collecting information that identifies individual readers. Aggregate figures tell us that a case page is popular; they do not tell us who read it.

We do not knowingly collect personal information from children. This site is a documentary archive intended for a general adult and older student readership. If you believe a child has sent us personal information through the contact form, write to us and we will delete it.

Depending on where you live, you may have rights over personal data relating to you, including the right to ask what we hold, to ask for a copy, to ask for correction, and to ask for deletion. Because we hold so little, these requests are usually simple to answer. Write to us and we will respond within the period required by the law that applies to you.

Our servers are located in North America, and information sent to this site is processed there. If you are visiting from elsewhere, your request necessarily crosses borders in order to reach us. By using the site you understand that this transfer takes place.

If we change this policy, the revised version will be posted here with a new date. Material changes will be noted plainly rather than buried. Questions about this policy should be sent through the contact page.

Cookie Policy

A cookie is a small text file that a website asks your browser to store, and that your browser sends back on later requests. Similar technologies, including local storage and pixel tags, do comparable work. This policy explains which of these are used on Anomaly Registry, by whom, and what you can do about them.

The registry itself is a reading site. It does not require a cookie to show you a record. We do not set cookies to track you across the internet, and we do not operate a first-party advertising identifier of our own.

Strictly necessary cookies may be set to keep the site secure and functioning, for example to help mitigate automated abuse of the contact form. These are limited in scope and are not used for advertising or profiling.

Advertising cookies are set by Google Ad Manager and by the advertising technology partners that operate within it. Their purposes include selecting which advertisement to show, capping the number of times a given advertisement is repeated, detecting invalid traffic and click fraud, and measuring whether an advertisement was actually seen.

Some of these advertising cookies support interest-based advertising, in which the advertisement you see is influenced by inferences drawn from browsing activity over time and across sites. Whether this occurs depends on your region, your settings, and the choices you have already made with those providers.

We may use aggregate analytics to count page views and understand which records are read. Where an analytics tool is used, we prefer configurations that do not identify individual readers and that do not follow readers across other websites.

We do not embed social media widgets, share buttons, or tracking pixels from social networks. This is a deliberate choice for an archive of this kind, and it removes an entire category of third-party tracking that most content sites carry.

Your browser gives you direct control over cookies. Every major browser can block third-party cookies, delete existing cookies, and clear stored data on exit. Blocking third-party cookies will not prevent you from reading anything on this site, though it may make the advertising you see less relevant.

You can also manage advertising preferences with Google directly through its advertising settings, and through the industry opt-out mechanisms operated in various regions. Because these are operated by third parties, we cannot make the choice on your behalf.

Where the law in your region requires consent before non-essential cookies are set, a consent notice will be shown and your choice will be respected. Withdrawing consent later is possible through the same mechanism.

Advertising partners may change over time. Rather than publish a list that goes stale, we identify the advertising system in use, which is Google Ad Manager, and direct you to Google's own disclosures for the current set of partners operating within it.

If you have a question about cookies on this site that this policy does not answer, write to us through the contact page and we will answer it.

Anomaly Registry is an independent documentary archive. It is published as a work of reference and public interest reporting, and it is supported by advertising. It is not affiliated with, endorsed by, or acting on behalf of any government, university, police service, coroner, scientific institution, or investigative body named in any record.

The register documents real events, including in some cases deaths, criminal investigations, and inquests. Where a record concerns a criminal matter, nothing on this site should be read as an allegation of guilt against any living person. Where an individual has been named in the public record, we report the fact of that naming and its evidentiary standing; we do not accuse.

Records are written from sources. Where a source is a primary document, we say so. Where a claim rests on secondary reporting, we say so. Where sources conflict, we state the conflict rather than quietly choosing a side. Where a widely repeated claim cannot be traced to any source, we say that too, because an untraceable claim is itself a finding.

Every record separates what is documented from what has been proposed. Hypotheses are labelled as hypotheses and attributed to whoever advanced them, together with their standing: peer-reviewed, contested, dismissed, or untested. We do not advance theories of our own.

The registry treats its own entries as provisional. Records carry a review date, and they are revised when the evidence changes. A record you read today may be corrected tomorrow, and that is a feature of the method rather than a defect in it.

Despite this method, errors will occur. If you find one, write to us. Corrections supported by evidence are made promptly, and we would rather be corrected than be consistent.

Scientific and forensic conclusions reported here are the conclusions of the researchers and officials who reached them, not ours. Reporting that a study proposed a mechanism is not an endorsement that the mechanism is correct, and the status assigned to a case reflects the state of the evidence rather than our opinion of it.

Nothing on this site is legal, medical, psychological, financial, or safety advice. Records that touch on medical or psychological matters, including mass psychogenic illness, report the diagnoses reached by clinicians at the time and afterward. They are not a substitute for professional care.

Quotations from sources are used sparingly and for the purposes of reporting, comment, criticism, and review. Copyright in quoted material remains with its owner. If you hold rights in material you believe has been used improperly, contact us and we will address it.

Original text, structure, record numbering, and editorial apparatus on this site are the property of Anomaly Registry. You are welcome to quote briefly and link to us. Wholesale reproduction of records, including reproduction for the purpose of training generative models, is not permitted without written permission.

Outbound links are provided as references. We do not control linked sites and are not responsible for their content, their accuracy, or their handling of your data.

This site is operated from British Columbia, Canada, and these notices are governed by the laws applicable there, without regard to conflict-of-law principles.

Accessibility Statement

Anomaly Registry is built to be usable by as many people as possible, including people who use screen readers, keyboard navigation, magnification, or other assistive technology. We treat accessibility as part of the build rather than as a later remediation.

We aim to meet the Web Content Accessibility Guidelines version 2.1 at level AA. This is a target we work toward continuously rather than a certification we claim to have completed.

The site is written as semantic HTML. Headings describe the real structure of a record, lists are marked up as lists, and the register is a genuine table with proper header cells, so that a screen reader can announce which column a value belongs to.

Every interactive element is reachable and operable by keyboard alone. Focus is visible at all times, and a skip link at the top of each page lets keyboard and screen reader users jump straight to the register without traversing the navigation.

Status is never conveyed by colour alone. Each status carries its written label in text, so a reader who does not perceive the colour difference between "Unsolved" and "Debunked" still receives the distinction.

Text colours have been chosen to meet or exceed the contrast ratios required at level AA against the page background. The typefaces are set at a comfortable reading size with generous line height, and the layout reflows to a single column on small screens without loss of content or function.

The site contains one animated element, an annotation on the home page. It respects the reduced-motion preference set in your operating system, and when that preference is on, the annotation appears without movement.

Text can be resized up to two hundred percent in the browser without breaking the layout or clipping content. Nothing on this site depends on a specific screen orientation.

Advertising is served by a third party. We control where advertisements are placed and label them as advertisements, but we do not control their internal markup, and their accessibility is ultimately determined by the advertiser.

We are developing an audio version of each record. When it is published, it will be an alternative way to receive the same content, not a replacement for the written entry, and the written entry will remain the authoritative version.

Accessibility work is never finished, and parts of this site will fall short of the standard we have set. If you encounter a barrier, we want to know about it. Please tell us what page you were on, what you were trying to do, and what assistive technology you were using.

Write to us through the contact page. We aim to respond within five business days and to fix genuine barriers promptly.

Terms of Use

These terms govern your use of anomalyregistry.com. By using the site you accept them. If you do not accept them, please do not use the site.

Anomaly Registry is offered as a free documentary reference. There is no account, no subscription, and no paywall, and nothing here is offered as a professional service of any kind.

You may read, quote briefly with attribution, link to, and print records for personal, educational, and research use. This is an archive, and it is meant to be used.

You may not reproduce records wholesale, republish the register or substantial parts of it, or scrape the site systematically. In particular, you may not reproduce this content for the purpose of training generative models. Permission for other uses can be requested through the contact page and is often granted.

Original text, the record numbering system, the status taxonomy, the editorial apparatus, and the design of this site are the property of Anomaly Registry. Third-party material quoted or referenced remains the property of its owners.

We work hard to make records accurate, and we correct them when they are not. Even so, the site is provided as it is, without warranty of any kind, express or implied, including any warranty of accuracy, completeness, or fitness for a particular purpose.

Records are provisional by design. A status can change, an identification can be confirmed or overturned, and a leading explanation can be displaced by a better one. Do not treat any record as a final word, and check the review date.

To the fullest extent permitted by law, Anomaly Registry and its operator are not liable for any loss or damage arising from your use of, or reliance on, anything published here.

If you send us a correction, a source, or a suggestion, you grant us permission to use it in the register. You will not be identified in connection with it unless you ask to be credited.

Do not use the contact form to send abuse, threats, spam, or automated submissions. We may block access to anyone who misuses the site or attempts to interfere with its operation.

Outbound links are provided as references and are not endorsements. We are not responsible for the content or conduct of any site we link to.

We may revise these terms. The version published here is the version in force. These terms are governed by the laws of British Columbia, Canada, and any dispute will be heard in the courts of that province.