Overview

Remote viewing asks a simple question: can a person accurately describe a place or object they have never seen and cannot reach by any normal means? The idea is ancient, but the experimental protocol is modern. In 1972, physicists Hal Puthoff and Russell Targ at SRI International began testing ordinary volunteers. They wanted to know if people could describe distant locations when given nothing but map coordinates or a randomly chosen target site. Their early results were published in Nature in 1974. The paper attracted both serious scientific interest and sharp criticism.³ The US government funded this work for over two decades under a classified program eventually called Stargate, which ran until 1995.²⁰

Remote viewing falls within the broader category of clairvoyance. That means the apparent perception of information about the physical world without using the known senses. What made the SRI protocol distinctive was its attempt to impose lab discipline on a notoriously slippery phenomenon. Targets were chosen randomly. Viewers were kept physically separated from the target. Transcripts were judged by people who had not seen the original session. These controls were designed to rule out the most obvious alternative explanations. For example, a viewer being told the answer, or a judge unconsciously matching descriptions to targets they already knew.

Why This Is Hard to Study

Remote viewing presents several specific challenges. These challenges make it genuinely difficult to study. They also mean that dismissing the entire literature is premature, even when individual studies fall short of ideal standards.

The first challenge is effect size. In science, an effect size tells you how large a difference is, not just whether it is detectable. Remote viewing effects are consistently small. A small effect is not the same as no effect. But it does mean you need many trials and many participants before you can be confident the result is real. Small effects are also the easiest to produce accidentally through subtle procedural flaws, which raises the stakes for careful methods.¹

The second challenge is that psi performance, if it exists, appears to be state-dependent. That means it may only show up under certain mental conditions, such as relaxed or meditative states. It may disappear under stress or skeptical scrutiny. State-dependent effects are hard to replicate because the relevant internal conditions are invisible to the experimenter.¹⁰

The third challenge is judging. Remote viewing produces free-form verbal or drawn descriptions. Turning those descriptions into a number requires a judge to compare them against a set of possible targets. The judge decides which one matches best. This process is vulnerable to subtle cues. If the judge knows anything about the session, the order of targets, or the viewer’s habits, that knowledge can bias the match even without any intent to cheat. Designing a truly blind judging procedure is harder than it sounds.¹²

A fourth challenge is the experimenter effect. Some researchers have reported that the same viewers perform better with some experimenters than others. Experimenter expectations may influence results in ways that are hard to control. If true, this makes replication across independent labs especially difficult.⁶

How the Experiments Work

A standard remote viewing trial works like this. Before the session begins, a set of possible targets is assembled. These might be photographs, geographical locations, or short video clips. One target is selected at random. The viewer sits in a separate room with no access to the target pool. They produce a verbal or written description and often a sketch. That transcript is then handed to a judge who has never seen the session. The judge compares the transcript against the actual target and several decoy targets. They rank them from best to worst match. A hit is scored when the actual target ranks first.

This design is called a blind judging procedure. The judge does not know which target was actually chosen. So their ranking cannot be influenced by knowing the answer. When neither the judge nor the viewer knows the answer during the session, the design is called double-blind. Double-blind designs are the gold standard. They close off the most common routes for information to leak into the result without any unusual process being involved.

The PEAR laboratory at Princeton ran a variation called precognitive remote perception. In this version, the viewer described the target before the agent had even traveled to the location. This added a time dimension: the viewer was not just describing a distant place but a future one.²

What the Evidence Shows

Any single remote viewing experiment gives a small, ambiguous result. So researchers combine many separate studies using a method called meta-analysis. The idea is straightforward: adding together enough studies makes even a small genuine effect statistically detectable, even if no single study is convincing on its own. The key question is whether the combined effect survives scrutiny about publication bias. Publication bias is the tendency for positive results to be published and negative results to be filed away and forgotten.

The first full meta-analysis of remote viewing research was published in 2023. It covered 36 studies conducted between 1974 and 2022. It found a small but consistent effect above chance. Tests for publication bias suggested the result was not simply an artifact of selective reporting.¹ An umbrella review published in 2021 combined 11 separate meta-analyses covering over 900 studies across 80 years. It found that remote viewing produced among the largest effect sizes of any anomalous cognition type studied, larger than ganzfeld or forced-choice designs.¹⁰

Remote viewing meta-analyses across multiple datasets show consistent effect sizes with confidence intervals excluding zero, and Bayesian analysis of the remote viewing database yields a Bayes factor in the billions in favor of a real effect over chance.¹⁷

A Bayesian analysis deserves a brief explanation. Standard statistics ask: how likely is this result if there is no real effect? Bayesian statistics ask a different question: given the data, how much more likely is the hypothesis that there is a real effect compared to the hypothesis that there is none? A Bayes factor of 1,000 means the data are 1,000 times more consistent with a real effect than with chance. A 2011 Bayesian meta-analysis of remote viewing data reported a Bayes factor in the billions. That means the data strongly favor a real effect over pure chance under that analysis.¹⁷

One intriguing finding concerns local sidereal time. Sidereal time is a clock based on Earth’s position relative to distant stars rather than the Sun. A 1997 analysis found that anomalous cognition effect sizes were roughly four times larger when trials occurred near 13.5 hours local sidereal time. This finding was replicated in an independent dataset.⁹ If confirmed, it would suggest that remote viewing performance is shaped by some physical factor related to Earth’s orientation in space. No mechanism has been identified.

It is important to separate the data from the interpretation. The data show a small, consistent statistical effect across many studies. The proposed mechanism, whether quantum entanglement, nonlocal consciousness, or something else, remains entirely contested and unestablished. Rejecting the proposed mechanism does not automatically explain away the data. Accepting the data does not commit anyone to any particular mechanism.¹⁹

Individual studies in the reference library

The studies listed below are individual remote viewing results currently held in the ESP-Nexus reference library beyond those already discussed above. What share of the published literature on remote viewing they represent has not been measured, so the table summarizes what the library holds rather than counting what has been published. The studies also report different kinds of number — direct-hit rates, sums of ranks, standardized effect sizes, z values and p values alone — and those cannot be added together into one bottom-line figure.

Both directions are represented in the table. The positive results span the founding SRI series — Price’s nine trials and Hammid’s nine, each significant by blind rank-order judging²⁶ — through the modern series: 14 of 36 ganzfeld-stimulated participants identifying the target photograph,²⁸ 38 of 48 associative trials predicting the DAX,³⁰ the ganzfeld condition of Roe and colleagues’ three experiments at 43 hits in 110 trials,³¹ 75 of 100 binary trials by five trained viewers,³³ the two applied Mobius fieldwork projects,²⁹³² and the believers group of the CIA follow-up study.³⁴ The null and below-chance results are fewer but direct: Wiseman and Watt’s four-trial mass-participation test performed close to chance, and the paper reports that its findings do not support the existence of remote viewing;²⁷ Kruth’s second judge scored significantly below chance on the target stock while the first did not differ from chance;⁷ Roe and colleagues’ waking condition was only marginally above chance, with direct-hit rates that stayed near the 25% chance level;³¹ and the first-group analysis of the CIA follow-up study was nonsignificant.³⁴

Several of the papers qualify their own results directly. Puthoff and Targ report that the learner pair’s series was not significant.²⁶ Wiseman and Watt note that with a small number of majority-vote trials there is a risk of a Type II error, and recommend a greater number of trials for future research.²⁷ Kruth describes his study as a pilot with a small sample size and a limited number of sessions, and reports that the sessions lost funds due to a complication in the investment process.⁷ Müller and Wittmann note that their present and future conditions were not run in balanced order, so the difference between them is potentially attributable to a learning or fatigue effect.³³ Schwartz and De Mattei present both a conservative and a less conservative chance calculation for the Leander site.³² Escolà-Gascón and coauthors, who describe their own approach as skeptical, report that emotional intelligence predicted between 9% and 19.5% of remote viewing scores and propose that emotions perceived during sessions may play an important role in the results.³⁴

The table deliberately excludes studies this page already treats in detail — the 2023 meta-analysis,¹ the PEAR remote perception database,² the 1974 Nature report,³ the Smith, Laham, and Moddel prediction series,⁴ the Firefly project,¹³ and the Mossbridge online tests¹⁴ — so that each row adds study-level results not already given above.

Skeptical Critiques and Discussion

Critique 1: The judging procedures in remote viewing studies leak cues, allowing above-chance scores without any unusual process.

Skeptic source: This critique was central to psychologist Ray Hyman’s dissent from the 1995 AIR review of the Stargate program; the canonical book-length presentation of the SRI cueing critique is Marks & Kammann (1980).²¹²²

Response: The 2023 meta-analysis applied publication bias tests and found the effect persisted across studies with tighter judging controls. The 2026 expert survey confirmed that blinding is now considered essential by practitioners, and that studies without it are not considered reliable.¹¹²

Analysis. Studies with cleaner methods still show an effect. But critics argue that no study has fully closed all judging artifact pathways, and the 2023 meta-analysis cannot fully separate clean from flawed studies in its pooled estimate.

Critique 2: The Stargate program produced no operationally useful intelligence despite decades of funding, suggesting the effect is too small or too unreliable to matter in practice.

Skeptic source: This argument was made by critics of the program following its 1995 declassification.²⁰

Response: Proponents distinguish between statistical significance in controlled experiments and operational reliability under field conditions. The 1995 AIR review acknowledged statistically significant lab effects even while questioning operational utility. Applied projects such as the Firefly currency-market study confirm that operational reliability remains undemonstrated at scale.¹³

Analysis. The large-scale Firefly project lost capital, and no applied remote viewing project has shown consistent above-chance returns under pre-specified conditions with an adequate sample size.

Critique 3: Meta-analytic results are inflated by publication bias and by pooling studies of widely varying quality.

Skeptic source: This is a standard critique applied to parapsychology meta-analyses broadly.¹⁹

Response: The 2023 meta-analysis applied three separate publication bias correction methods (3PSM, RoBMA, and trim-and-fill) and found the effect survived all three. The 2011 Bayesian meta-analysis found no consistent negative correlation between effect size and sample size. Such a correlation would be expected if publication bias were the primary driver.¹¹⁷

Analysis. Publication bias tests are imperfect and cannot rule out selective reporting entirely. The effect survives current best-practice tests, but the field lacks a large pre-registered adversarial replication that would settle the question.

Critique 4: Remote viewing functions inside Faraday cages and at ocean depths, so it cannot operate via any known electromagnetic signal, making a physical mechanism implausible.

Skeptic source: Skeptics have used this as an argument against the phenomenon; proponents have turned it around.¹¹

Response: Williams (2019) notes that remote viewing’s apparent insensitivity to electromagnetic shielding actually undermines common objections based on signal transmission. Those objections assume information travels like a radio wave. The absence of a known mechanism is not the same as evidence against the phenomenon. But it does mean no proposed mechanism has been confirmed.¹¹

Analysis. The absence of a mechanism does not explain away statistically significant results, but it does mean the phenomenon lacks grounding in established physics.

Open Questions: What Would Settle This

The remote viewing literature has accumulated enough positive results that the central question is no longer simply whether any effect exists. The more pressing questions are whether the effect survives fully adversarial pre-registered designs, whether it is large enough to be useful in practice, and what factors moderate it.

The most decisive study would be a large pre-registered adversarial replication. Pre-registered means the researchers publicly commit to their hypotheses, sample size, and analysis plan before collecting any data. This prevents the common problem of exploratory analyses being reported as if they were confirmatory. Adversarially designed means skeptics and proponents jointly agree on the protocol before it runs. So neither side can later claim the design was unfair. Such a study would need enough trials to detect the small effect sizes reported in the meta-analyses. That probably means several hundred trials across multiple independent viewers. Judging would need to be conducted by a team that includes at least one committed skeptic.¹¹²

A second open question concerns the sidereal time effect. If anomalous cognition performance genuinely peaks near 13.5 hours local sidereal time, experiments timed to that window should show reliably larger effects. A pre-registered study that randomly assigned trial timing to on-peak and off-peak windows, with the timing concealed from viewers and judges, would test this directly. A null result would substantially weaken the sidereal time hypothesis. A positive result would be one of the most striking findings in the field.⁹

A third open question is whether the judge matters as much as the viewer. The 2021 ARV study found that one judge performed significantly below chance while another performed at chance on the same viewer transcripts. This suggests the judge is a critical variable that has been largely ignored.⁷ A study that randomly assigned multiple judges to the same transcripts, with judges varying in their attitude toward psi and their experience, would clarify whether judge selection is a confound in the existing literature or a genuine moderator of the effect.

Finally, the question of practical utility remains open. Small studies have reported striking short-run accuracy in financial prediction. Large studies have not. A pre-registered multi-site study with a fixed investment strategy, a pre-specified stopping rule, and independent auditing of all trades would provide a definitive test. The Firefly project came closest to this design but lacked full pre-registration.¹³

Modern context

Remote-viewing research does not exist in isolation. Three bodies of mainstream science inform how to read its evidence base: the formal statistical framework used to score blind-judged tasks, the cognitive psychology of mental imagery and how people distinguish imagined content from perceived content, and the broader literature on how experimenter expectations bias outcomes. None of these fields treat remote viewing as a hypothesis worth testing in their own right, but each constrains how to interpret remote-viewing results.

The blind-judging procedure used in SRI, PEAR, and post-Stargate remote-viewing studies is, in statistical terms, a free-response signal-detection task. Signal-detection theory, formalized by Green and Swets in the 1960s, treats every judging decision as a noisy classification between signal (the actual target) and noise (the decoy pool). The framework predicts that hit rates depend not just on any underlying signal but on the size and similarity of the target pool, the judge’s decision criterion, and how granular the rating scale is. Translating remote-viewing outcomes into SDT terms is rare in the parapsychology literature but would let researchers compare effect sizes across protocols on a common metric.²³

A second mainstream-science constraint comes from research on mental imagery and source monitoring. People routinely confuse the source of their own mental content: imagined details get re-categorized as remembered details, and vivid imagery can feel as definite as perception. Johnson and colleagues’ source-monitoring framework documents these errors in non-parapsychological populations performing standard cognitive tasks. In a remote-viewing session, the viewer produces imagery they did not request and must decide what is signal and what is noise from their own mind — exactly the kind of decision the source-monitoring literature shows is error-prone even for ordinary perception.²⁴

Third, the experimenter-expectancy literature established by Rosenthal and Rubin documents that researchers’ beliefs about likely outcomes subtly bias every step of a study, from instructions given to subjects through the rating of free-response data. Their meta-analysis of 345 studies across eight domains found a mean expectancy effect size of d = 0.70 — large enough to account for the kind of small-but-consistent effects reported in remote-viewing meta-analyses if blinding is incomplete. Modern remote-viewing protocols address this with double-blinded judging procedures, but the historical SRI work and most pre-1995 studies do not meet that standard.²⁵