← Back to The Haunt Report
Field GuideNov 15, 2026 · 10 min read

How Does a Spirit Box Work? The Ghost Hunter's Radio Scanner, Explained

If the EMF meter is what a modern ghost hunt looks like, the spirit box is what it sounds like: a fast, stuttering rasp of static, with scraps of voices surfacing and going under again several times a second, and somebody in the dark saying 'did you hear that?'

It is also the newest tool in the standard kit. The Ouija board has a patent from 1891 and the EMF meter was borrowed from electricians, but the spirit box was invented for this purpose, by a known person, within living memory, and most of its history can be dated to the month.

This is the field guide to it. What the box does electrically, where it came from, why nearly everyone who listens to one hears words, and how to run a session whose results somebody else can check.

What is a spirit box?

A spirit box, also called a ghost box, is a radio receiver modified or built to scan across the AM or FM band continuously without ever stopping on a station. It plays a fraction of a second of whatever is on each frequency, then moves to the next. Investigators ask questions aloud and listen for words in the resulting stream, on the theory that a spirit can use the fragments of audio to form a reply.

There is no component in it that detects anything other than radio. Everything that comes out of the speaker went in through the antenna or was generated by the circuit itself.

How does a spirit box work?

Take the most common model, the P-SB7, and the FM band. In the United States the FM broadcast band runs from 88 to 108 megahertz and is divided by regulation into 100 channels of 200 kilohertz each, from 88.1 up to 107.9. The SB7 steps through them one at a time, forwards or backwards, pausing on each for a period you choose between 100 and 350 milliseconds. At the fastest of those settings a full pass of the band takes about ten seconds; at the slowest, about thirty-five. Its sellers also describe a synthetic white noise that the box inserts between the steps.

Now compare that with speech. Conversational English runs at roughly four to six syllables a second, and a typical syllable lasts somewhere between 150 and 250 milliseconds. So each stop on an occupied channel hands the listener about one syllable, often less: half a word from a presenter, a sung vowel, the front edge of an advert, a burst of guitar. Each stop on an empty channel hands over a slice of hiss.

What reaches your ears is therefore a rapid string of unrelated syllables and noise, in different voices, at different volumes, several per second. That is the raw material. Nobody disputes this part, including the people who make the boxes. The dispute is over what assembles the syllables into words.

Edison, Raudive and a magazine from 1995

Sellers of spirit boxes like to begin the story with Thomas Edison, and there is something to it. In October 1920 Edison gave two interviews, one to B. C. Forbes in The American Magazine and one to Austin C. Lescarboura in Scientific American, published on 30 October, in which he talked about an apparatus that personalities who had left this life might be able to operate. He was careful to say he was promising nothing. The Edison Papers project at Rutgers, which has catalogued his surviving records, reports that no verifiable evidence of any work on such a device has ever been found. Edison supplied the idea and nothing else.

The practical ancestors are the tape-recorder experimenters covered in our beginner's guide to EVP: Friedrich Jürgenson in Sweden and Konstantin Raudive, whose book Breakthrough appeared in English in 1971. Among Raudive's methods was recording from a radio tuned to an unused frequency, on the reasoning that the static gave the voices something to work with. That reasoning is the whole theory of the spirit box, a generation early.

There was also one famous attempt at a two-way machine. On 6 April 1982 George Meek, of the Metascience Foundation in Franklin, North Carolina, held a press conference at the National Press Club in Washington to announce Spiricom, a set of tone generators through which his colleague William O'Neil had, Meek said, held more than twenty hours of conversation with a dead scientist. Meek gave the plans away. Nobody else managed to reproduce the result, and Meek's own explanation was that it needed O'Neil's mediumship. A 1987 article in Fate magazine argued that the spirit's buzzing voice was what an electrolarynx sounds like.

The direct line starts later and smaller. The October 1995 issue of Popular Electronics carried a cover story called 'Ghost Voices', by an author who signed himself Konstantinos, with the cover line 'Investigating Electronic Voice Phenomena'. It summarised Jürgenson and Raudive and included a circuit for a white-noise generator to use in place of an empty radio frequency.

One of its readers was Frank Sumption, a hobbyist in Littleton, Colorado. In 2002 he built the first of what everyone else came to call Frank's Boxes. The early design used a white-noise source to drive a random voltage, and the voltage to drive the tuner from a car stereo, so that it hopped about the AM band unpredictably. In 2007 he added a steady sweep with an adjustable rate, which is the design every later box copied. He played the output into a chamber with a microphone in it, which he called an echo box, and recorded the result.

Sumption did not sell his boxes. He gave them away and published the plans. A Skeptical Inquirer article in 2010 put the count at more than sixty and tested number 63; the obituaries in August 2014 put the total at about 180. He believed the voices came from spirits and from other entities besides, he said that he had no means of testing the idea, and he maintained that the results depended on the listener rather than on the device. That last point is the one this article will keep returning to.

The Shack Hack and the P-SB7

Frank's Boxes were scarce, so in late 2007 hobbyists found a shortcut. Several cheap RadioShack portable radios with a scan button, the 12-469, 12-470 and 12-587 chief among them, silence their speaker while they hunt for the next station. Cut the wire or the circuit-board trace that carries the mute signal and you hear every step of the scan instead. This was the Shack Hack, and until a commercial unit arrived it was how most people got a ghost box. Accounts differ on who worked it out first.

The first purpose-built commercial unit came from Gary Galka of D.A.S. Distribution in East Granby, Connecticut. Galka's seventeen-year-old daughter Melissa died in September 2004 after a car accident, and he began designing equipment for paranormal investigators in the years that followed; one of his meters, the Mel-8704, is named for her and for the years of her birth and death. The P-SB7 was tested and named with the help of the investigator Chris Fleming, was shown on a live Ghost Adventures broadcast in October 2009 according to its retailers, and was on general sale by early 2010. The same programme devoted an episode to the family, 'The Galka Family', first shown on 20 April 2012.

The SB7 costs a little under a hundred dollars and has been the default ever since. It is worth knowing how it came to exist. It was not designed as a cynical product. It was designed by a bereaved father who wanted it to work.

Why do you hear words in a spirit box?

Because the human hearing system fills in speech that is not there, and has been shown to do so under controlled conditions for more than fifty years.

In 1970 the psychologist Richard Warren played twenty students a recorded sentence in which 120 milliseconds, one consonant and its edges, had been cut out and replaced by a cough. Nineteen of the twenty reported that no speech sound was missing, and the one who reported a gap named the wrong sound. Nobody could say correctly where in the sentence the cough had fallen. When Warren replaced the sound with silence instead of noise, listeners noticed the hole at once. The brain restores speech that is masked by noise, and does not restore it across silence. A spirit box supplies a burst of noise between every fragment, and its shortest step is 100 milliseconds.

In 1981 Robert Remez and colleagues reduced a spoken sentence to three whistling tones. Listeners told nothing mostly heard whistles, electronic music or interference. Listeners told it was speech could often write the sentence down. What you expect to hear changes what you hear, from the same sound.

The closest study to the ghost hunter's situation is by Michael Nees and Charlotte Phillips, published in Applied Cognitive Psychology in 2015. They played people clips of supposed EVP taken from a ghost-hunting television show, mixed in with clear speech, speech buried in noise and plain noise. Half the participants were told the study was about paranormal voices and half that it was about hearing speech in noise. Those given the paranormal framing reported a voice in the EVP clips on 48 per cent of trials; the others on 34 per cent. More telling is what happened when people who did hear a voice were asked what it said. For real speech they agreed with one another 95 per cent of the time. For the EVP clips they agreed 13 per cent of the time, and they matched the television investigators' own reading less than 1 per cent of the time. A later attempt to repeat the framing effect reproduced it only in part, but the agreement figure is the one that matters here: if the words were in the recording, listeners would agree on them.

As for direct tests, in a study published in 2001 the psychologist Imants Barušs recorded just over sixty hours of radios tuned between stations while assistants invited the dead to speak. He found a handful of voice-like sounds and concluded that none was clearly anomalous. When the Skeptical Inquirer writer Karen Stollznow and an engineer tested Frank's Box number 63 in 2010, they got unintelligible noise. We could not find a peer-reviewed controlled experiment on a spirit box as such, in either direction, and anyone who tells you the device has been scientifically validated or scientifically disproved is overstating what has been done.

What is the Estes Method?

The most interesting development since the SB7 is a procedure, not a gadget. In January 2016 three investigators working at the Stanley Hotel in Estes Park, Colorado, Karl Pfeiffer, Connor Randall and Michelle Tate, tried splitting the session in two. One person, the receiver, wears a blindfold and sound-isolating headphones plugged into the spirit box, and says aloud any word they hear in the sweep. The others ask the questions, which the receiver is not supposed to be able to hear. If the receiver's words answer questions they never heard, the reasoning goes, expectation has been removed. The method took its name from the town and reached a wide audience through the 2019 documentary series Hellier.

The instinct is exactly right. It is the same instinct as blindfolding the sitters at a Ouija board: take the information away from the person producing the answer, and see whether the answer survives.

The execution is where it falls short. Kenny Biddle, writing in Skeptical Inquirer in April 2019, tried it and found he could still hear his wife talking through isolating headphones with the spirit box running, particularly in the gaps of static. Headphones of that kind take roughly 25 decibels off outside sound, and conversation happens at 60 to 70. The receiver is also still doing the thing the science above describes, choosing words out of syllables. And the questioners, who can hear both sides, decide afterwards which of the receiver's words counted as answers. One of the two leaks has been narrowed and the other is untouched.

How to run a spirit box session worth keeping

Start with the advice of the people who built the thing. Chris Fleming's own notes for the SB7, written in 2010, warn the user that 'you will be receiving at times, bits and pieces of radio broadcast information', and recommend keeping sessions to fifteen or twenty minutes because after that the mind starts looking for anything that might be a response. That is the man who helped test and name the device telling you the two main ways it will fool you. Believe him.

Record the entire session on a separate recorder, in one continuous file, and say the sweep rate, the band and the direction out loud at the start. Write your questions down beforehand and read them from the list. A question composed on the spot tends to be shaped by the last thing the box seemed to say.

Ask questions that have a specific answer you can check and that nobody could hit by accident. 'Is anyone here?' will be answered yes by any vowel. 'What year did you die?' needs four particular syllables in order.

Run a control. Take the same box, the same settings and the same list of questions to somewhere with no reputation at all, your kitchen will do, and run it for the same length of time. Count the responses you get there. Whatever the number is, that is your baseline, and it will not be zero.

Then do the step that almost nobody does. Cut out each clip you think is a response and play it to three people who were not there, without telling them what you heard or what was asked. Have each write down, separately, what the voice says. If they agree with one another and with you, you have something. If you get three different answers, you have the Nees and Phillips result, and you have saved yourself from posting it.

The claim worth chasing is a narrow one: a reply of several syllables, in one continuous voice, that runs across several steps of the sweep. A single station cannot produce that, because the box has already left it. If you ever record one, keep the whole unedited file, because the first thing anyone will ask is what came before it.

The part worth filing

Frank Sumption said that the result depended on the listener and not on the box. He meant it as a claim about sensitivity. It is also a fair summary of what the experiments show. The spirit box is a machine for producing ambiguous syllables at high speed, and a person listening to one is a machine for turning ambiguous syllables into words. Put the two in a dark building with a story attached and words will arrive, every time, for everyone.

That does not make the session worthless. It makes the session a test of whether you can tell your own pattern-finding from something else, and that test can be passed or failed on the record. So if you file a spirit box report on Spookr, file it like any other evidence: the location and time, the settings, the question exactly as written, the original recording, your kitchen baseline, and what three people who were not in the room heard when you played it back. Stay safe, stay spooky.

Got a story of your own?

Log your sightings, attach your evidence, and put your haunt on the map.

Get Spookr