Why do we dream? Global database offers new clues

Dreams have always fascinated scientists, yet studying them has been a logistical nightmare. Each lab used its own methods, making it nearly impossible to compare results or draw broad conclusions.

Now, researchers have unified more than 2,600 carefully timed awakenings into the largest dream dataset ever created.

The new Dream EEG and Mentation Database (DREAM), hosted by Monash University, links patterns of brain activity before people wake with what they report experiencing.

By transforming decades of scattered sleep studies into a single standardized resource, the project finally gives scientists a way to test long-standing theories about when and how dreams arise.

Bringing clarity to dream reports

Dream science has long suffered from incompatible methods. Labs used different questionnaires, free-form reports, and even different definitions of what counts as a dream.

The new database resolves that by coding each awakening into three categories: “experience” (specific content recalled), “experience without recall” (knowing you dreamed but lacking details – so-called white dreams), and “no experience.”

That simple framework lets researchers compare nightmares to nap dreams, lucid dreams to ordinary ones, and clinical groups to healthy volunteers without talking past each other.

Two decades of dream data

The entries span the early-2000s to 2024 and include overnight studies with repeated awakenings, first-thing-in-the-morning reports, and daytime naps.

One dataset pairs EEG with magnetoencephalography for a richer view of brain dynamics. Participants range from children with developmental dyslexia to older adults, frequent sleep talkers, and trained lucid dreamers.

All of it is packaged in a consistent file structure so analysis pipelines can run across studies without bespoke reformatting.

Dreams across every sleep stage

From 1,550 awakenings with full sleep staging and reports, clear patterns emerge. Light non-REM (N1) produces dream reports most often, at 88 percent.

N2 yields reports in a little over half of awakenings. Deep non-REM drops to roughly half, and REM – though famous for vivid dreams – returns reports about 81 percent of the time.

The upshot: conscious experiences appear throughout sleep, not just in REM, with rates in non-REM hovering around 40 to 60 percent depending on depth and timing.

Clean signals, clear comparisons

To be included, recordings needed at least two standard EEG leads, a minimum of 20 seconds of continuous sleep before the wake cue, and sampling at 100 Hz or higher.

Files arrive raw or minimally preprocessed, then pass automated and manual checks for structure, metadata consistency, and artifacts.

Any record with more than 10 seconds of noise in the final 20 seconds before awakening is excluded.

The result is a set of polysomnography files in a shared format, each paired with sleep stage, dream classification, demographics when available, and the exact experimental context.

Tracking the brain’s dream markers

To showcase what the scale enables, the team re-scored several high quality datasets with automatic sleep algorithms, matching expert human labels above 72 percent accuracy.

Importantly, the algorithms output probabilities, not just single-stage stamps. Non-REM awakenings that included reports of experience tilted toward more wake-like probability patterns than non-REM without reports. This suggests measurable shifts in brain state when consciousness is present.

The experts also tested direct dream prediction from EEG. Using both standard spectral features and more advanced signal measures, classifiers separated dreaming from non-dreaming better than chance in both REM and non-REM.

In REM, the best approach reached about 70 percent accuracy. These early results point toward objective markers of a subjective state – signals that can now be refined on a much larger canvas.

Dreams and consciousness connected

Why are we conscious in some moments of sleep but blank in others? Why do some people recall dreams nightly while others almost never do? What sets routine lucid dreamers apart?

The database is designed to probe these questions head-on. It also opens avenues beyond sleep.

Methods that reliably detect awareness in dreaming brains could translate to monitoring depth of anesthesia or assessing covert consciousness in disorders of consciousness, where behavior is absent but awareness may persist.

Openness with guardrails

Not every dataset is identical in access. Some are fully open; others require permission from the original investigators, in line with ethics approvals.

But all share standardized metadata that researchers can search and browse without hauling gigabytes of signals.

That lowers the barrier for hypothesis generation and lets teams assemble targeted subsets – for example, all N2 “experience without recall” awakenings in young adults, or all lucid dream eye signals.

The future of dream discovery

The initial release covers 2,643 awakenings – a starting point meant to grow. A volunteer core team spanning Australia, France, Sweden, the UK, and more curates contributions and keeps the standards tight.

The project dovetails with a broader push toward open, reproducible neuroscience, echoing the impact of community datasets in other domains where shared resources have sped discovery.

Dream research has long been data-poor and method-fragmented. With a shared taxonomy, rigorous quality gates, and thousands of aligned awakenings, the DREAM database turns isolated studies into a collective instrument.

The database shows that dreaming is widespread across sleep, offers first objective markers to chase, and gives the field a practical way to test theories about consciousness.

As more labs contribute, the fog around what the sleeping brain is doing – and when it is doing it – should finally start to lift.

The full study is published in the journal Nature Communications.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–