The Horse That Passed Every Test
In 1904 Berlin, a horse could add, subtract, spell, and tell time by tapping its hoof — and the scientists sent to catch the trick couldn't find one.
In 1888, a retired Berlin schoolmaster named Wilhelm von Osten bought a horse. He was not looking for a pet or a workhorse. Von Osten, who had spent years teaching mathematics to unenthusiastic schoolboys, had a theory that animals were far smarter than anyone gave them credit for, and that the right teacher could prove it. He had already tried the experiment once, on a cat and a bear, with no success. The horse — a good-natured Russian trotter he named Hans — was his third attempt, and the one he was determined to get right.
His earlier attempts had gone nowhere. Von Osten had first tried the experiment on a cat, which showed no interest whatsoever in being taught, and then on a bear, which was reportedly too dangerous to work with closely for long. A horse, patient and food-motivated, turned out to be a far better subject.
He started small, the way he had with children: tapping out the number one, then asking Hans to tap once in return. He rewarded correct answers with carrots and bread, and simply withheld the reward when Hans got it wrong, over and over, for years. By the early 1890s, Hans could tap out any number up to ten. Von Osten kept going. He taught Hans to add, subtract, multiply, and divide by tapping the answer with his right hoof. He taught him fractions. He taught him to recognize playing cards, tell time from a clock face, identify musical notes, and pick out colors. Using a code where each letter of the alphabet corresponded to a number of taps, von Osten claimed Hans could even spell out answers to questions, one laborious tap-sequence at a time.
None of this happened quickly. Von Osten spent well over a decade on Hans’s education, in daily sessions, long before the horse ever performed for a paying — or, in this case, non-paying — public. By the time Berliners started lining up to watch, what they were seeing was the end product of years of unpublicized, repetitive drilling that nobody outside von Osten’s courtyard had witnessed.
By 1904, Wilhelm von Osten was giving free public demonstrations in a courtyard in Berlin, and Clever Hans — as the newspapers had taken to calling him — had become a sensation. Crowds gathered daily. Visitors would call out arithmetic problems, and the horse would tap out the correct response with what looked, to everyone watching, like genuine comprehension. Someone would ask Hans what three plus five was, and after a pause, he would lift his right hoof and tap eight times, then stop. Ask him to find the square root of a number, and he’d get that too. The horse appeared to understand spoken and written German, to do arithmetic better than most of the children von Osten used to teach, and to know it.
Von Osten never charged admission and never claimed Hans was a trick act. He genuinely believed he had demonstrated something important about animal intelligence, and he wanted scientists to confirm it. That turned out to be a fateful decision, because scientists took him up on it — and they did not go easy on him.
The commission
The horse’s growing fame reached Germany’s board of education, which in September 1904 convened a panel to settle the question of whether von Osten was running a hoax. It became known as the Hans Commission, and its thirteen members were deliberately chosen to include people who knew how performing animals were usually faked: a circus manager, a cavalry officer experienced with horses, a veterinarian, the director of the Berlin Zoological Gardens, several schoolteachers, and Carl Stumpf, a prominent philosopher and psychologist who chaired the group.
The commission put Hans through extensive testing. They asked questions von Osten hadn’t prepared. They brought in strangers to pose questions instead of von Osten, to rule out a secret signal between trainer and horse. They checked for hidden wires, subtle sounds, anything mechanical. They found nothing. In its report, the commission concluded that there was no trickery of any kind involved — no rigged apparatus, no coded signals von Osten was deliberately sending. As far as thirteen experienced professionals could determine, the horse really was doing what it appeared to be doing.
That finding only made the story bigger. A horse that could do arithmetic, verified by an official scientific panel, was a far more remarkable headline than a horse that could do a magic trick. The case made international news, including coverage in The New York Times.
But the commission had only ruled out deliberate fraud. It hadn’t explained how the horse was actually getting the answers right, and it passed that harder question on to one of its members for closer study: Carl Stumpf’s young research assistant, a man named Oskar Pfungst.
The second look
Pfungst was not a horse expert. He was a psychologist, and he approached Clever Hans less like an animal trainer sizing up a stunt and more like an experimentalist trying to isolate a variable. Instead of asking whether Hans could do math, he asked a narrower, stranger question: under exactly which conditions did Hans get the answer right, and under which conditions did he fail?
He started changing the setup in small, controlled ways. He had Hans answer questions when the person asking didn’t know the answer themselves. He had questioners stand farther away, or out of the horse’s line of sight entirely. He fitted Hans with blinders that restricted what he could see directly in front of him.
The pattern that emerged was stark. When the questioner stood where Hans could see them and knew the answer to the question, Hans performed just as well as he had for the commission — correct almost every time. But when the questioner didn’t know the answer, Hans’s accuracy collapsed, often to barely better than chance. And when Hans couldn’t see the person asking the question at all, his performance collapsed too, regardless of whether that person knew the answer. The horse needed two things to get a right answer: a human in view who knew what the right answer was, and nothing blocking his view of them. Take away either one, and the arithmetic genius vanished.
That told Pfungst the horse wasn’t retrieving the answer from inside his own head. He was getting it from the person standing in front of him — and the person wasn’t doing it on purpose.
This was, in its own way, a more interesting result than if Pfungst had simply found a hidden signal and ended the investigation there. A deliberate secret cue would have made Hans an ordinary trained animal performing an ordinary trick, the kind circus acts had used for decades. What Pfungst was describing instead was a horse that had become sensitive to something nobody — not von Osten, not the commission, not the audiences who had watched Hans for years — had known they were producing.
Pfungst began watching the questioners instead of the horse. He noticed that whenever someone asked Hans a question requiring a hoof-tap answer, they would unconsciously lean forward slightly and tilt their head down a fraction, anticipating the moment Hans would start tapping — the same way a person watching someone count out change might lean in without thinking about it. As Hans’s taps approached the correct number, the tension in the posture of everyone watching would subtly build: a held breath, a slight stiffening, a microscopic shift in expression. The instant Hans reached the right number of taps, that tension broke — the watchers relaxed, however faintly, because the suspense was over. Hans had simply learned, through years of reward and correction, to read that release of tension as his cue to stop tapping. He wasn’t counting to eight. He was watching a roomful of people until they stopped looking like they were waiting for something, and then he quit tapping.
It was an almost absurdly subtle signal — a change in posture lasting a fraction of a second, invisible to the people producing it, and most of them flatly denied doing anything at all when Pfungst told them what he’d found. To confirm it wasn’t a fluke of a single observer’s body language, Pfungst tested it on himself: he practiced asking Hans questions while deliberately controlling his own posture, and then tried deliberately (if subtly) cueing wrong answers, and could reproduce the effect on command. Whoever stood in front of Hans was broadcasting the right number of taps without knowing it, and Hans had become an extraordinarily sensitive instrument for picking that signal up.
Not a hoax, something stranger
The distinction mattered enormously, and it’s the reason the story outlived the horse. Von Osten had not built a secret trick. He hadn’t trained Hans to respond to a deliberate, hidden hand signal the way a carnival act might. He had, without realizing it, spent years training a horse to become exceptionally good at reading minute, involuntary human body language — and had convinced himself, along with a commission of professional skeptics, that he was watching arithmetic instead.
Von Osten reportedly never accepted Pfungst’s explanation and continued giving demonstrations until his death in 1909. Ownership of Hans then passed to Karl Krall, a jeweler and fellow believer in animal reasoning, who moved the horse to Elberfeld and kept presenting him as proof of equine intelligence, alongside other “talking” horses he trained using similar methods. Hans was eventually requisitioned as a military horse during the First World War. What became of him after that was never recorded; popular claims that he was killed in action or eaten by soldiers are widely repeated but have never been substantiated by any source closer to the events.
Pfungst published his full findings in 1907, and the phenomenon he’d identified acquired a name of its own: the Clever Hans effect, now standard terminology in experimental psychology and animal cognition research for any situation where a human or animal subject appears to demonstrate a capability, but is in fact picking up unconscious cues from an observer who already knows the “correct” response. It became a foundational cautionary tale in the design of scientific experiments, part of the reasoning behind double-blind study design — if the person recording results doesn’t know which subjects got the real treatment and which got the placebo, they can’t unconsciously signal the expected answer, the way von Osten and his audiences had signaled Hans.
The effect has turned up since in far more consequential settings than a Berlin courtyard. Studies of drug-detection dogs have found handlers can unintentionally cue their dogs toward a location where the handler — wrongly — believes contraband is hidden, producing false alerts that mirror exactly what Pfungst documented in 1904. Facilitated communication, a discredited technique once used with nonverbal autistic people in which a facilitator physically supported the person’s hand over a keyboard, was shown through controlled testing to often reflect the unconscious movements of the facilitator rather than the communication of the person being “assisted” — a Clever Hans problem with much higher stakes than a math trick. More recently, researchers studying AI systems have invoked the same name for models that appear to solve a task correctly while actually having learned to exploit incidental patterns in how the test itself was constructed, rather than the underlying skill the test was meant to measure — a benchmark score that looks like competence but is really just a very sensitive detector of shortcuts the test designers didn’t know they’d left lying around.
In each of these later cases, the structure of the original story repeats almost exactly: a subject appears to demonstrate real understanding, the people evaluating it are acting in good faith, and the actual explanation turns out to be that the evaluation itself was leaking the answer through some channel nobody had thought to control for. Pfungst’s contribution wasn’t just catching one horse. It was showing that this particular failure mode — call it, after him, a Clever Hans problem — could happen to careful people doing careful work, which is exactly why it keeps recurring in fields having nothing to do with horses.
What makes the Clever Hans case endure isn’t that a horse turned out to be a fraud. By every account, nobody involved was lying. Von Osten seems to have genuinely believed what he was demonstrating; the commission genuinely found no trick; the audiences genuinely saw a horse get the right answer, over and over, because it was reading them with a precision most of them would never have thought themselves capable of projecting. The uncomfortable finding wasn’t about the horse’s limits. It was about how easily careful, well-intentioned, skeptical humans could build an elaborate and sincere belief on a signal none of them knew they were sending.