Science and precognition meet at a difficult question: can a person obtain information about a future event without ordinary clues or inference? Some researchers have reported results they interpret that way. Other studies have failed to reproduce important findings, and disputes about methods and analysis remain.
The evidence discussed below does not establish precognition as a reliable ability. That conclusion is different from saying that every unusual experience is dishonest or meaningless. A personal experience can deserve a careful hearing without becoming proof that information traveled backward through time.
Quick Guide
- Distinguish a claim of precognition from ordinary prediction.
- Read positive findings alongside replication attempts and methodological criticism.
- Above-chance averages require analysis; “above 50%” alone is not proof.
- Physiological anticipation is a research question, not a personal warning system.
- Keep dreams and hunches separate from decisions requiring verified information.
Define the Claim Before Evaluating It
A weather forecast predicts conditions using observations and models. An experienced person may anticipate a familiar outcome without consciously listing every clue. Precognition makes a different claim: information about an event is available before it occurs without those ordinary routes.
“Premonition” is often used for a vague anticipatory feeling. “Retrocognition” refers to claimed knowledge of an otherwise unknown past event. These labels describe proposed experiences; they do not establish the abilities or explain how they would work.
Before calling a result precognition, ask whether the information was genuinely unavailable, how the future event was selected, and whether the response was securely recorded before that selection. An unexpected match is the beginning of an investigation, not its conclusion.
Why a Compelling Dream Story Is Not Enough
Historical accounts can contain detailed scenes, named witnesses, and sincere recollections. Those features may make a report worth investigating, but they do not automatically exclude ordinary explanations. A witness remembering that someone mentioned a dream is not the same as an independently preserved, detailed record made before the event.
For any dream report, ask:
- Is the original record available, with a credible date?
- Which details were written before the event, rather than added afterward?
- How many dreams did not appear to match anything?
- Could the person have encountered relevant information earlier?
- How broadly are words such as “journey,” “danger,” or “meeting” interpreted?
- Were witnesses independent, or did they discuss the story before recording it?
A collection of striking stories still needs answers to those questions. A count of cases judged impressive by a researcher is not, by itself, a calculation of the probability that precognition exists.
How Laboratory Experiments Approach the Question
One approach uses forced choices: a participant selects an option, then a computer selects or reveals a target. If there are two equally likely target positions selected independently, the chance expectation for a guess is one half. Other designs have different chance expectations.
Another approach reverses the order of a familiar psychological procedure, asking whether a later event is associated with an earlier response. These tasks can involve different measures, so findings from one task should not be presented as though they automatically confirm every other task.
Strong designs need secure timing, appropriate randomization, prevention of ordinary information leakage, clear scoring, and a plan for analyzing the data. They also need replication: can the result appear again under comparable controls?
Selected Findings and Replication Results
This is a focused evidence review, not a comprehensive systematic review of every precognition paper. The sources are linked so readers can examine what each study actually tested.
Bem’s 2011 Experiments
Daryl Bem’s 2011 paper reported nine experiments designed around time-reversed psychological effects, with statistically significant main results in eight. One task asked participants to choose where an image would appear; another examined whether later word practice related to earlier recall. These reports prompted methodological debate and replication efforts. They should be described as reported findings, not established future-seeing powers.
Three Unsuccessful Recall Replications in 2012
Ritchie, Wiseman, and French reported three preregistered, independent attempts to reproduce Bem’s ninth experiment, involving retroactive facilitation of recall. Each used 50 participants. None reproduced the claimed effect. This is evidence about that particular recall task; it is not a test of every possible dream or intuition claim.
A Positive Meta-Analysis in 2015
Bem and colleagues’ meta-analysis combined 90 experiments from 33 laboratories in 14 countries. It reported an overall effect size of Hedges’ g = 0.09, and 0.06 for independent replications after excluding Bem’s original experiments. The authors argued that their analyses supported an anomalous effect and were not substantially explained by selection bias. Those are their conclusions; pooling studies does not eliminate the need to examine their designs and later replications.
A Large Replication Project Reported in 2025
Walleczek and colleagues’ 2025 project tested the effect claimed in Bem’s first experiment, using three studies and 26,483 participants. Study 1 did not reproduce the predicted positive effect. Exploratory analysis suggested a result in the opposite direction; Study 2 reproduced that opposite-direction result, but Study 3 did not reproduce it.
The authors judged a method-derived anomaly, such as a false-positive finding, more plausible than a psi explanation under conventional standards. It would be misleading to report only Study 2 as confirmation of precognition—or to describe the project as three identical null results. The sequence and direction of the findings matter.
Why “Above 50%” Does Not Mean “Not Random”
Under an independent, balanced two-option guessing model, half is the long-run expectation, not a requirement that every finite run contain exactly equal numbers of hits and misses. Random variation can produce an above-half result.
The size of the sample changes how much information a percentage provides. A small excess in a short run is not equivalent to the same percentage across a large, properly controlled study. Even a statistically unusual result needs checks for ordinary clues, analysis choices, and measurement problems.
Consider four separate questions:
- Effect size: How large is the observed difference, not just whether it crosses a significance threshold?
- Uncertainty: How precisely is that difference estimated?
- Analysis plan: Was the test selected before seeing the results, or chosen from many alternatives?
- Replication: Does the result recur with new data and adequate controls?
A small p-value is not the probability that precognition is real. It evaluates data under a specified statistical model and assumptions. Testing many outcomes, stopping when a favorable result appears, or reporting only successful runs can distort the picture.
Presentiment: Does the Body React Before the Stimulus?
Presentiment research examines physiological activity before stimuli that are intended to be unpredictable. Mossbridge, Tressoldi, and Utts’ 2012 meta-analysis examined 26 reports and described a small anticipatory effect. Measures included electrodermal activity and heart rate. The authors called for replications across laboratories and stated that the cause remained to be determined.
In a 2014 methodological commentary, D. Samuel Schwarzkopf questioned factors including unequal stimulus probabilities, expectation, trial order, and possible analysis artifacts from earlier responses. These are proposed methodological concerns, not a demonstration that every reported dataset has the same flaw.
The practical distinction is important: an unexplained group-level difference in a laboratory measure does not establish that a sensation in your chest predicts a phone call or danger. It also does not justify ignoring a bodily symptom or assuming it is a message.
What Makes Evidence More Informative?
On a small screen, scroll the table sideways to read every column.
| Question | More informative approach | What remains uncertain |
|---|---|---|
| When was the claim recorded? | A fixed record before target selection or outcome knowledge. | Whether ordinary information was available. |
| How was it scored? | A predefined rule that allows a genuine miss. | Whether the rule answers the intended question. |
| Were all attempts included? | Complete results, including null findings and exclusions. | Whether missing studies or data affect the conclusion. |
| Was the result repeated? | Independent, adequately controlled replication. | How broadly a result generalizes beyond its task. |
Preregistration means specifying a research plan before examining the relevant data and submitting it to a registry. It helps distinguish planned tests from later exploration. It does not repair every design flaw or guarantee that a result will replicate.
A Personal Log Is Not a Laboratory Study
If you want to examine your own impressions, keep the activity low-stakes. Before checking the outcome, record the date, exact impression, specific prediction, deadline, scoring rule, prior information, and confidence. Later add the actual result and alternative explanations.
“A major change is coming” has no clear failure condition. “My evening language class will be canceled by Thursday at noon” does. Even then, knowing that the teacher was unwell would be relevant prior information. Do not remove it from the record because the prediction felt sudden.
This notebook exercise is not formal preregistration, and it does not control all sources of bias. Its modest purpose is to make your interpretation more transparent, not to certify an ability.
Worked Example: Do Not Move the Deadline
The following example is fictional. Owen dreams of an empty classroom and writes on Monday that Thursday’s language class will be canceled by noon that day. He notes that the teacher had mentioned a possible scheduling problem.
The class proceeds as planned. On Friday, a different workshop is canceled. Owen records the original claim as a miss; he does not substitute the other event. He also records the prior scheduling information, which could have influenced the dream.
The dream may still be interesting to reflect on. What it does not supply is a successful prediction under the rule Owen wrote in advance. Keeping that distinction is more informative than collecting only apparent matches.
Does Precognition Mean the Future Is Fixed?
The studies reviewed here do not establish that events are predetermined, that emotional events travel backward through time, or that all psychic labels describe one mechanism. Those are additional philosophical or theoretical claims, not conclusions that follow from a disputed statistical result.
Telepathy, clairvoyance, and precognition also make different claims about what information is accessed. A surprising match does not reveal its mechanism merely because several labels could be applied. Definitions and controls remain important.
For everyday decisions, letting intuition guide you can mean noticing a concern and checking it. Curiosity about precognition is compatible with withholding certainty and relying on verified information when the stakes matter.
Frequently Asked Questions
Has science proved precognition?
The studies discussed here do not establish precognition as a reliable ability. Some experiments and meta-analyses report positive findings, while important replication attempts have failed and methodological questions remain.
How is precognition different from prediction?
Prediction can use known patterns, experience, and available information. Precognition is the claim that information about a future event is obtained without ordinary access or inference. A surprising correct guess does not by itself distinguish them.
Does guessing correctly more than half the time prove an effect?
No. The chance expectation depends on the task, and observed results vary around it. Sample size, independence, the analysis chosen in advance, multiple testing, and replication all matter.
What is presentiment?
Presentiment research examines whether physiological measurements before an unpredictable stimulus differ according to what appears afterward. Reported differences require scrutiny of expectation, timing, measurement, and analysis; they are not proof that your body can reliably warn you about future events.
Does a dream that comes true count as scientific evidence?
A dated record can make a report easier to examine, but one apparent match does not rule out chance, prior information, flexible interpretation, or selective recall. Misses and the exact wording recorded before the event matter too.
What does preregistration add?
It records the research plan before the relevant data are examined and distinguishes planned tests from later exploration. It improves transparency but does not automatically make a design sound or a result correct.
Should I use a precognitive feeling for an important decision?
Treat it as a reason to check relevant information, not as a verified instruction. Use evidence, consent, and appropriate professional advice for health, safety, money, or relationship decisions.