In Berlin in 1904, a horse called Hans appeared able to answer questions that went far beyond ordinary training. Wilhelm von Osten, his owner, believed that education had developed the animal’s capacity to understand numbers and other problems. Hans usually gave numerical answers by tapping a hoof. He could also perform for people other than his owner, which made a simple explanation based on a private signal between one man and his horse seem inadequate.
The investigation by Oskar Pfungst, working with Carl Stumpf and Erich von Hornbostel, established a different mechanism. The horse’s success depended on whether a person knew the answer and whether Hans could see that person. Pfungst then identified and deliberately reproduced the small bodily movements that started or stopped the tapping. The documented explanation concerns Hans’s demonstrations. It does not require the conclusion that horses lack intelligence, or that every successful questioner knowingly deceived an audience.
A performance that deserved testing
Stumpf’s introduction to Pfungst’s account makes clear that the explanation was not apparent from casual observation. Hans had attracted experienced observers, and successful performances in the owner’s absence challenged the assumption of a straightforward conjuring trick. Stumpf described his own uncertainty about how movements unnoticed by human spectators could guide the animal. His initial possibilities included very slight sounds as well as bodily gestures.
The early committee’s report of 12 September 1904 was therefore not the final experimental explanation. Stumpf presents that stage as recognition that the case merited investigation. Subsequent tests asked what happened when nobody present knew the correct answer. More detailed work ran from 13 October to 29 November, with Pfungst conducting experiments and von Hornbostel keeping records. A report dated 9 December set out the resulting explanation.
The later book and its English translation preserve the distinction between those stages. The 1904 investigation is the incident period recorded here; 1907 and 1911 belong to the German work and English edition. Quoting the first favourable impression without the later controls would misrepresent the sequence. Equally, treating the final explanation as something that every original visitor should instantly have recognised misses the difficulty the investigators themselves described.
Changing who knew the answer
Pfungst’s first major comparison was between trials in which the questioner knew the answer and trials in which the answer was hidden from the people present. In numeral tests, cards were placed so that Hans could see them while the human participants could not. Other trials allowed the questioner to know which numeral was displayed. Alternating the conditions helped address objections that the horse simply happened to be tired during failures or was disturbed by an unfamiliar arrangement.
The report gives 49 numeral trials without human knowledge of the answer and 42 with it. Correct responses were reported as 8 per cent in the first condition and 98 per cent in the second. These are the investigator’s published results, rather than a modern reanalysis of an independently available dataset. Their significance lies in the large change when the information available to the human was changed.
Word tests produced the same pattern. Words were arranged so that the questioner did not initially know their positions. Twelve trials without that knowledge produced no correct answers; fourteen with knowledge produced correct answers throughout. Hans’s apparent reading depended on information possessed by the questioner, rather than only on the writing available to the horse. This supported a cueing explanation, but did not yet identify the sensory channel.
What the blinders showed
The investigators next restricted the horse’s view. This was not as simple as fitting any eye covering and counting successes. Hans tried to obtain a view around obstacles, and small or poorly positioned blinders left uncertainty about what he could see. Pfungst explicitly separated trials in which the questioner was seen, definitely unseen or of uncertain visibility.
With sufficiently large blinders, the report describes 102 trials: 35 in which the questioner was definitely unseen, 56 in which the questioner was seen, and eleven uncertain. The corresponding success rates were 6, 89 and 18 per cent. When the questioner stepped into the visible field and repeated a problem, performance could recover. This comparison helped distinguish loss of visual information from the mere discomfort or novelty of wearing the equipment.
Earlier tests with smaller blinders had given less clear results. Pfungst discussed how counting uncertain trials as if vision had certainly been blocked could create the impression that sight was unnecessary. This detail matters because claims that Hans succeeded while blindfolded were not automatically fraudulent reports. Some observers may genuinely have believed that an inadequate obstacle excluded the relevant information.
Speech was also unnecessary for some successful performances. Pfungst could obtain the expected tapping without speaking the number aloud. The horse’s dependence on human knowledge therefore did not demonstrate telepathy: the questioner could be communicating information through an ordinary sensory route without recognising the communication.
Finding the movements
Observation identified a recurring sequence. A questioner bent the head and upper body slightly forward, after which Hans began tapping. At the desired number, a small upward movement of the head signalled the end. The horse drew his foot back in a larger movement. A subsequent, more conspicuous straightening of the person’s posture could distract an observer from the earlier and smaller effective signal.
Pfungst reported these movements in von Osten and other successful questioners, with individual variations in their size and clarity. People who produced late, premature or absent movements obtained corresponding errors. The important step was then experimental: suppressing the movements disrupted the response, while deliberately supplying them allowed Pfungst to control it. He could obtain the animal’s reactions without asking the supposedly relevant question.
The investigators also considered whether the human movement merely followed the horse’s stopping rather than caused it. Their timing arrangement used two observers with stopwatches, one watching the questioner and the other the animal. This was an attempt to examine temporal order, with limitations inherent in human observation and the watches available. It shows that the account was not confined to noticing a plausible gesture and assuming that it explained everything.
Why this was not a finding of deliberate fraud
The December report stated that questioners could give useful signs without intending to do so. Schillings and Pfungst obtained correct replies before they were aware of their movements. Once Pfungst had learned the mechanism, intentional movements could reproduce what had previously happened involuntarily. That distinction allows the same visible performance to arise through different degrees of human awareness.
The report specifically resisted treating this as sufficient evidence that von Osten had constructed an elaborate deception. It proposed that the horse had gradually learned to attend to changes in posture associated with the desired answer, reinforced by rewards such as bread and carrots. The owner’s intention to teach arithmetic could therefore produce a different and unintended learning outcome.
Some of Pfungst’s wider conclusions about the horse’s senses, emotions and mental life extend beyond the direct demonstration of cueing. The secure case finding is narrower: the tested numerical and reading performances depended on human knowledge and visible movements, and the relevant responses could be reproduced by supplying those movements. The experiment did not establish a general ceiling on animal cognition.
Assessment, image and location
This case is classified as explained because the investigation documented a cause and demonstrated control of the responses. Hans’s apparent independent problem solving in the tested performances was accounted for by visual cueing. Neither uncertainty about the owner’s intentions nor the horse’s impressive sensitivity requires leaving that particular mechanism unidentified.
The historical photograph shows Hans before a crowd, with Schillings presenting him according to its source description. It is contextual evidence of the demonstrations, not a photograph of a specific controlled trial or an image that by itself proves the cueing mechanism. The reproduced image comes from Karl Krall’s 1912 book and is identified as public domain by its file record; the later publication date should not be substituted for the investigation date.
The map uses a broad Berlin reference at 52.517, 13.383, rounded from the sourced city coordinates. It does not identify the original courtyard, a surviving stable, a private address or a present-day place open for investigation. The physical setting was important to the experiments, but publicising an exact residential site is unnecessary to understand their results.