Overview

The word telepathy was coined in the 1880s to name a simple idea: one mind reaching another across distance with no physical link. In the lab, this means a setup with two people. A sender concentrates on a target, like a picture or a film clip. A receiver, isolated in another room, tries to describe what the sender is experiencing.

This sender-receiver structure is what distinguishes telepathy from clairvoyance. In clairvoyance there is no sender at all. Someone simply tries to perceive a hidden object directly. Telepathy assumes an active mind on the transmitting end.¹¹

Telepathy is one of the most commonly reported psychic experiences in everyday life. People say they thought of a friend just before that friend called. Identical twins report sensing each other’s distress.¹⁰ Anecdotes like these are vivid but impossible to test, because we cannot rule out coincidence, shared knowledge, or selective memory. So researchers moved the question into the laboratory, where targets are random and conditions are controlled.

Why This Is Hard to Study

If telepathy exists at all, it is faint. The effects reported in careful experiments are small. That single fact shapes every difficulty that follows.

A small effect is easy to drown in noise. To detect it reliably you need many trials, large samples, and clean procedures. Any single study tends to give an ambiguous result. To pull a small effect out of the data, researchers combine many studies using a method called meta-analysis. When you add that much data together, a small real effect can become statistically detectable even if no single study is impressive on its own. The trouble is that meta-analysis also amplifies any flaws shared across studies.

The phenomenon also depends on conditions that are hard to control. Who the sender and receiver are, how close they feel, whether they believe in psi, and even their cultural background all seem to move results.⁹ When the effect itself varies with mood and relationship, a clean repeat run becomes very difficult. That is part of why a result that appears in one lab may shrink or vanish in another.

How Telepathy Has Been Tested

Three main experimental families have carried the telepathy question. Each one tries to solve a specific problem with the earlier ones.

The first was dream telepathy. At the Maimonides Dream Lab in Brooklyn, running from the early 1960s into the late 1970s, a sender in a distant room concentrated on a randomly chosen picture while a receiver slept. Dreams happen mostly during a stage of sleep called REM, short for Rapid Eye Movement, when the eyes flick back and forth under closed lids. Scientists watched for that flicker using small electrodes near the eyes, so they knew exactly when the sleeper was dreaming. They woke the receiver during REM and recorded the dream before it faded, then later checked whether the dream matched the sender’s target.¹³

The second family is the ganzfeld. The word is German for “whole field.” The receiver lies in mild sensory uniformity: halved ping-pong balls over the eyes under red light, soft static in headphones. The idea is that with normal senses quieted, a faint telepathic signal might surface. Meanwhile a sender in another room views a randomly selected target image or film. Afterward the receiver picks which of several images best matches their experience.¹

The third family is newer and technological. One 2020 study replaced the ganzfeld’s quiet room with virtual reality, immersing the sender in vivid 30-second VR clips of exciting activities. The thought was that a stronger sender experience might transmit more. Receivers then ranked candidate images.¹¹ A separate strand of work uses real-time communication with sleepers, where experimenters and people in REM sleep exchanged simple signals.⁸

What the Experiments Found

The headline result comes from the ganzfeld. When Honorton and the psychologist Daryl Bem combined the autoganzfeld trials in 1994, receivers picked the correct target more often than blind guessing would predict.² The excess was modest but, across the full database, statistically significant.

Across the autoganzfeld database, receivers identified the correct target about 32% of the time, against a chance expectation of 25%.²

The picture darkens when later studies are added. A 1999 meta-analysis of ganzfeld studies done after Honorton’s reported a smaller or null effect. A 2010 re-analysis that folded in more recent work re-found a small effect, especially in studies using selected, experienced participants.⁸ So the answer you get depends heavily on the date range and the inclusion rules.

The related distant-intention paradigms tell a steadier story. Here one person tries to affect another’s attention or physiology from a distance, rather than send a picture. Three independent meta-analyses of these tasks landed on nearly the same small effect size.⁹ The convergence is notable because the three tasks were run by different labs on different continents.

The newer technological tests have not produced clean wins. The virtual-reality telepathy study did not meet its main goal.¹¹ Its preregistered hit-rate measure stayed at chance. An exploratory analysis of receivers’ top two ranked choices looked suggestive, but per ESP-Nexus’s evidential standard an exploratory finding after a preregistered failure is hypothesis-generating only. It does not count as positive evidence until an independent preregistered replication confirms it.

Both directions are represented in the set. The telephone and email program’s own reports include the 2003 telephone tests, where the 40% overall success rate rose to 53% with familiar callers and fell to exactly chance with unfamiliar ones,¹⁵ the 2005 email series at 43% against a 25% chance expectation (z = 9.49),¹⁶ the automated online test at 29.3% against 25%,¹⁷ the email and SMS tests at 41.8% and 37.9% against a 33.3% chance level,¹⁸¹⁹ the mobile-phone tests at 41.8% with three callers and 55.2% with two,²³ and the rapid online test’s much smaller margin of 26.7% against 25%.²⁰ Outside that program, Wahbeh, Cannard, Radin, and Delorme report 48.1% against a 33.3% chance expectation on their telepathic trials,²⁴ and Silberstein and Bigelow report a significant anomalous-interaction effect in five monozygotic twin pairs.²⁶ The chance-level results come from replications by other groups: Schmidt, Erath, Ivanova, and Walach’s studies with self-selected samples,²¹ the Zener-card conceptual replication by Rey and colleagues,²² and Stedall and Tressoldi’s preregistered two-caller test.²⁷

Several of the papers qualify their own results directly. Sheldrake and Stedall report that their continuous “conference call” tests showed no significant telepathic effects, with the positive result coming from tests in which trials were conducted less frequently and non-callers were unaware that trials were taking place.²⁵ Schmidt and coauthors conclude that they “could not find any anomalous cognition effect in self-selected samples” while adding that their data “strongly suggest that there are a few participants who are able to score reliably and repeatedly above chance.”²¹ The 2025 meta-analysis reports that hit rates under precognitive conditions were at chance, that selected participants had significantly higher hit rates than unselected participants, and that hit rates were significantly higher when callers and participants shared an emotional bond.²⁸ Stedall and Tressoldi’s null result carries its own exploratory observation, that intuitive responders outperformed rational ones, offered as a lead rather than a conclusion.²⁷

How Researchers Interpret the Data

It is worth separating the data from the explanations. The data are the hit rates and effect sizes above. The explanations are a different and far more contested matter.

The traditional interpretation takes telepathy literally: some signal or connection carries information from sender to receiver. But this reading runs into a problem. Shielding the rooms does not seem to block the effect, which is strange for any ordinary signal.⁷ That puzzle has driven some theorists to abandon the idea of a sender entirely.

So even among researchers who accept that the data show something, there is open disagreement about whether “telepathy” names anything real. The proposed mechanisms are plural and contested. The experimental finding of a small above-chance effect can stand on its own without committing to any one of them.

Skeptical Critiques and Discussion

Critique 1: The original ganzfeld database had flaws in how the studies were run.

The psychologist Ray Hyman’s 1985 critique identified problems in the early ganzfeld work, including possible sensory leakage and weaknesses in randomization and judging.²

Honorton’s response was not to argue but to rebuild the method. The autoganzfeld automated target selection with an electronic random source, isolated sender and receiver, and kept the judging blind, directly addressing the named flaws.²

Analysis. Hyman’s critique and Honorton’s redesign are often cited as a model of how a skeptical challenge can improve a paradigm. The reforms eliminated several specific artifacts. Whether the residual effect in the cleaned-up database is real remains the point still contested by later meta-analyses.

Critique 2: Newer ganzfeld studies do not replicate the original effect.

A 1999 meta-analysis by Milton and Wiseman of post-Honorton studies reported a smaller or null effect, suggesting the result does not reliably repeat.⁸

A 2010 re-analysis by Storm, Tressoldi, and Di Risio added more recent studies and again found a small effect, strongest in studies using selected, experienced participants.⁸

Analysis. The two meta-analyses cover overlapping but different sets of studies and reach different conclusions, which shows how sensitive the answer is to inclusion rules and participant selection. Storm and Tressoldi’s later work continued to report evidence of anomalous communication using the ganzfeld with selected participants.⁸ An independent, large, preregistered ganzfeld series with fixed inclusion rules has not resolved the disagreement.

Critique 3: Sophisticated psi paradigms fail when others try to replicate them with adequate power.

Walach and colleagues ran two preregistered replications of the matrix experiment, a psychokinesis paradigm, and both failed to reproduce the original positive findings even though the studies had enough power to detect the original effect.¹⁴

The same authors argue the failure is not proof that nothing is there. They propose a “no-signal-transfer” model in which non-local correlations cannot be turned into a reliably extractable signal, predicting exactly this decline.¹⁴

Analysis. The matrix experiment tests psychokinesis, not telepathy directly, but the replication problem it documents recurs across psi paradigms. The authors’ post-hoc power analysis (88% and 69%) refutes the claim that underpowering caused the nulls. Their interpretation that the failure supports a non-local model is itself untestable in the usual way, which the authors acknowledge.

Critique 4: The large telephone-telepathy effects come from one research program, and replications by other groups have found chance results.

Stedall and Tressoldi’s preregistered replication asked 117 participants to identify which of two callers was calling; the 294 correct identifications in 604 trials (48.6%) fell slightly below the 50% chance level (z = −.61, p = .76).²⁷ Schmidt, Erath, Ivanova, and Walach’s three replication studies found 26.7% correct against a 25% chance expectation in the first study (not significant), and the authors conclude that they “could not find any anomalous cognition effect in self-selected samples.”²¹ A German conceptual replication using Zener cards recorded 493 telepathy hits where 480 were expected by chance (p = .26).²²

The Sheldrake, Stedall, and Tressoldi meta-analysis of 26 telecommunication telepathy experiments reports a standardized effect size of 0.17 (p = 1 × 10⁻⁷) and no significant difference between the results of Sheldrake and his colleagues, who carried out most of the studies, and independent replications; it also reports that hit rates were significantly higher when callers and participants shared an emotional bond.²⁸ Schmidt’s own series ended with one pre-selected participant identifying 24 of 60 callers correctly (p = .007),²¹ and the 2024 Sheldrake–Stedall tests produced a 57% hit rate against a 50% chance level in the 266 trials where non-callers were unaware that trials were taking place.²⁵

Analysis. The split runs along design and participant lines. Automated tests with emotionally bonded callers report large above-chance hit rates,¹⁵¹⁶²³ while the two-caller preregistered replication²⁷ and the card-guessing conceptual replication²² report chance results. Schmidt’s conclusion names both directions in one breath: no anomalous cognition effect in self-selected samples, alongside data suggesting a few participants who score reliably above chance.²¹

Open Questions: What Would Settle This

The telepathy debate has stalled on a single recurring weakness: the effect, if real, is small and unstable, and meta-analyses disagree because they count different studies. The fix is not another clever paradigm. It is a large, preregistered, adversarial study with rules locked in advance.

A decisive experiment would look like this. Skeptics and proponents agree on the protocol before any data are collected: the ganzfeld procedure, the participant selection criteria, the randomization source, the blind judging method, the sample size, and the analysis plan, all fixed and published in advance. The sample would be large enough to detect the small effect that distant-intention meta-analyses predict, which means hundreds of sessions, not dozens.¹¹ No exploratory analyses would be allowed to count as confirmation.

Two specific questions could be settled this way. First, does the selected-participant effect (the claim that experienced or believing receivers do better) survive a preregistered test, or is it an artifact of optional stopping and flexible scoring?⁸ Second, does the sender matter at all? A clean comparison of sender versus no-sender conditions, run blind and at scale, would test the very thing that defines telepathy.¹ If the effect appears only when a sender is present, that supports the literal reading. If it appears equally without a sender, the informational-psi theorists are right that telepathy is not a separate category.⁷ Either result would be progress. So far, no such adversarial study at adequate scale has been published.