Overview

A precognitive dream is a dream whose content seems to match an event that happens afterward. The claim is not just that the dream resembled the future. The claim is that the match is too close to be a coincidence, and that the dreamer had no normal way to know what was coming.

People have reported these dreams for thousands of years. Surveys of the general public find that roughly a third of people say they have had at least one experience they would call precognitive.² Dreams are by far the most common setting people describe.⁴ That gap, between how common the experience feels and how thin the laboratory evidence is, is the central puzzle of this topic.

The serious laboratory work began at the Maimonides Dream Laboratory in Brooklyn, New York. From the early 1960s through 1972, Montague Ullman and Stanley Krippner ran controlled dream-ESP studies there. Most of that work tested dream telepathy. A smaller set of studies, with a single gifted participant named Malcolm Bessent, tested dream precognition specifically.

Why This Is Hard to Study

Dream precognition is hard to study for reasons that go beyond the usual difficulty of testing a faint effect. The very nature of dreams works against clean measurement.

First, dreams are vague, shifting, and emotional. A single night’s dream can be read to match dozens of different future events. The looser the dream, the easier it is to find a future that fits.

Second, human memory makes the problem worse. People remember the rare dream that seemed to come true and forget the thousands that did not. This is a memory bias, not a paranormal signal, and it works on believers and skeptics alike.²

Third, even in a controlled study, the effect appears to be small. When researchers combine many studies, the typical effect size is modest, on the order of a small correlation.⁸ A small effect needs large samples to detect reliably. Dream studies are expensive and slow, so most are small, and small studies give noisy, unstable results.

How the Experiments Work

Modern dream precognition experiments follow a simple but strict order of events. The dreamer records a dream. Only after the dream is locked in does a random process pick a target. The dreamer, or an outside judge, then compares the dream against the real target and several decoys.

That ordering matters enormously. If the target is chosen by chance only after the dream is fixed, then no normal information could have leaked into the dream. There is nothing to perceive yet, no sender to read, and no way to cheat. That is what makes the design a test of precognition specifically, rather than of any other channel.

A recent home-dream study shows the modern, low-cost version of this design. Participants slept at home, recorded their dreams, then rated a future target image and three decoys for similarity to what they had dreamed. The target was selected at random only after the ratings were in. Because everything happened online and the target was chosen afterward, there was no opportunity for hidden cues or leakage.¹

What the Studies Found

The honest summary is that results are mixed and small. Some controlled studies find a small effect above chance. Others find nothing. The same researchers have produced both kinds of result.

Caroline Watt, a parapsychologist who has approached the topic from a skeptical, methods-first angle, ran an online dream precognition study that found no link between people’s dreams and the future target clip.³ A separate circadian study tested whether dreaming at the time of peak melatonin (around 3am) would boost precognition. It did not. The overall hit rate barely cleared chance and was not significant.⁴

A reanalysis weighting studies by sample size found that the largest dream-ESP studies produced a near-zero effect, while only the smallest studies produced a notable one.⁹

A 2025 case study revived the single-gifted-dreamer approach with new technology. A skilled lucid dreamer recorded dreams, targets were randomly chosen afterward, and the matches were scored by artificial intelligence text-comparison models as well as by human judges. The study was preregistered, meaning the analysis plan was fixed in advance to prevent cherry-picking.⁵

Individual studies in the reference library

The studies listed below are the individual dream-precognition results currently held in the ESP-Nexus reference library beyond those already discussed above. What share of the published literature on dream precognition 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 counts, hit rates, effect sizes, critical ratios and p values alone — and those cannot be added together into one bottom-line figure.

Both directions are present in the table. The strongest positive results are the two Maimonides series with Malcolm Bessent, each returning five direct hits in eight nights,¹⁰¹¹ the Sargent and Harley free-response series at a 41% direct-hit rate,¹² the two pooled estimates — r = 0.33 for the Maimonides set¹³ and a Stouffer Z of 5.32 across the fifty-study set¹⁹ — and Watt’s planned analysis of 200 home-dream trials at a 32% hit rate against 25% chance.¹⁶ The null results include Robinson’s covert test, in which 100 participants selected the target-matched description 52% of the time against a 50% chance level,¹⁴ the circadian study, in which both the dream task and the forced-choice task ended non-significant,¹⁵ and the registry study, in which prospective precognitive dreams shared no phenomenological marker.¹⁸

Several of the papers qualify their own results directly. Watt and Valášek report that the hit rate of Watt (2014) is inflated by the omission of the dropouts’ data, and that the excess hits in that study did not appear to be attributable to the participants’ dreams resembling the targets more than the decoys, because the judges’ ratings of targets were not significantly higher.¹⁶¹⁷ Luke and colleagues place their own result against the Sherwood and Roe review, in which the three post-Maimonides precognition studies fared the worst of the set, with effect sizes ranging from r = −.34 to .07.¹⁵ Storm and coauthors report that improvements in study quality were not related to effect size, but that effect size declined over the 51-year period.¹⁹ Sherwood and Roe note that the greater success of the Maimonides studies may be due to procedural differences, including that the later receivers tended to sleep at home and were generally not deliberately awakened from REM sleep.¹³

What Researchers Make of It

It is important to keep the data separate from the explanations. The data are some small, fragile positive effects and a lot of nulls. The explanations for any real effect are many, contested, and far more speculative than the data warrant.

Some researchers propose that the dreaming mind is a naturally altered state that is more open to anomalous information. On this view, sleep quiets ordinary sensory noise and lets a faint signal through.¹ This is a proposed mechanism, not an established one.

Others reach for physics. One theoretical paper argues that precognition would only make sense if the future is fixed rather than open, and tries to ground that in relativity and time-symmetric quantum mechanics.⁶ These are conceptual proposals. They do not rest on dream data and remain firmly contested.

Skeptical Critiques and Discussion

Critique 1: Apparent precognitive dreams are produced by ordinary memory bias, not anomalous knowledge.

Watt and colleagues argue that people selectively remember dreams that seem to come true and forget those that do not, and that they actively hunt for matches between unrelated events.²

In a controlled study, participants recalled roughly two to three times as many confirming dream-event pairs as disconfirming ones from a constructed diary, and this bias appeared in believers, agnostics, and skeptics alike.² Prospective registries, where dreams are logged before any event, find very few precognitive-looking dreams, in sharp contrast to the high rate of casual reports.²

Analysis. Watt’s two studies identify specific, measurable non-psi mechanisms that inflate spontaneous reports. These bear on everyday accounts collected after the fact. They do not directly address the laboratory designs where the target is randomly chosen after the dream is recorded, since those designs remove memory bias by construction.

Critique 2: The pooled dream-ESP effect is an artifact of small studies and publication bias.

Howard reanalyzed a dream-ESP database with sample-size weighting and found the effect shrank by about two-thirds and lost significance after correcting for missing studies.⁹

A broader free-response meta-analysis reports a small but persistent positive effect across decades with no decline, and finds no significant difference between precognition and other channels.⁸

Analysis. Storm reports a small surviving effect across the broad free-response literature; Howard reports that within the dream-ESP subset the effect collapses once large studies are weighted appropriately and missing studies are imputed. The two analyses use different databases and different weighting choices, and both publish their methods. Precognition specifically came out near chance in Howard’s reanalysis.

Critique 3: Popular and personal-journal claims overstate the evidence and misattribute methods.

A review of a popular book on precognition found it devoted only a few pages to actual experiments, leaned heavily on a single 1971 Bessent study, and omitted failed projects.⁷

The controlled home-dream and AI-scored studies, by contrast, preregister or randomize their procedures and report their nulls and cautions openly.¹

Analysis. Katz documents specific omissions and misattributions in one popular treatment. This is a critique of science communication rather than of the controlled experiments themselves. The peer-reviewed dream studies that survive scrutiny are the small, carefully randomized ones, not the anecdote-driven popular accounts.

Critique 4: The Maimonides dream experiments were dismissed in mainstream psychology reviews as methodologically unsound.

Child’s review in American Psychologist documents the critical treatments: Hansel’s 1980 revision argued that sensory cues permitted by the Maimonides procedures explained the results, and gave more weight to the two negative replication attempts at the University of Wyoming than to the sum of the Maimonides research; Alcock’s 1981 volume wrote that a separate control group “would appear essential”; the Marks and Kammann book gave the dream experiments no mention at all.²⁰

Child reports that in several reviews the experiments “have been so severely distorted as to give an entirely erroneous impression of how they were conducted”: Hansel’s sensory-cue interpretation rested on a misreading that Akers (1984) showed to be erroneous, and the studies used within-subject controls parallel to those of standard psychological research rather than lacking controls, as Alcock implied.²⁰

Analysis. Child’s paper documents misdescription in secondary reviews rather than delivering a verdict on the phenomenon: he writes that the accumulating evidence for such anomalies is not totally convincing, and the two Wyoming replication attempts he discusses did return negative outcomes, which is the record Hansel weighted most heavily. Child’s complaint is that psychologists guided by the critical reviews are prevented from gaining accurate information about how the experiments were actually conducted.²⁰

Open Questions: What Would Settle This

The path forward is not more small studies. It is a few large, decisive ones. A preregistered home-dream experiment with several hundred to a few thousand participants would have the statistical power that current studies lack. The 2024 home-dream design is cheap enough to scale, and its target is chosen randomly after the dream is recorded, so leakage is impossible.¹

An adversarial collaboration would help most. A believer and a skeptic could agree in advance on the design, the sample size, the analysis, and the threshold for success, then run it together. If a large preregistered study with sample-size-appropriate analysis still found an effect, the publication-bias explanation would weaken sharply. If it found nothing, the small positive results would look like artifacts.

The AI-scoring approach is worth scaling too. Automated text-comparison judging removes human judging bias and can be applied identically across thousands of trials. A large preregistered study using AI scoring, with the embedding models fixed in advance, could test whether the single-dreamer signal generalizes or vanishes.⁵ Until such a study is run and replicated, dream precognition remains a small, fragile, and openly disputed effect.