Why Do We Dream? What the Neuroscience Says

Every night, your brain builds a world, populates it with people, drops you into a story with no clear beginning, and then mostly deletes the file before you wake up. That’s dreaming, and despite being one of the most universal human experiences, it’s also one of the least understood. There’s no shortage of theories. There’s just not much agreement on which one is right, or whether dreaming needs a “reason” at all.

So let’s look at what the current research actually supports: how dreams are generated in the brain, and the leading ideas about what they might be for.

Dreaming isn’t tied to just one sleep stage

The old assumption was simple: REM sleep equals dreaming, other sleep stages don’t. That’s been out of date for a while now. Dreaming also occurs during non-REM (NREM) sleep, and late-night NREM awakenings can produce dream reports that are essentially indistinguishable from REM dreams [2]. What differs is the character of the dream. NREM dreams tend to be shorter, more conceptual, less bizarre, less emotionally charged, and more tied to whatever you were actually thinking about that day, compared to the longer, more narrative, more visually strange dreams typical of REM sleep [2,3].

This creates a real puzzle for neuroscience. REM and NREM sleep have almost entirely different EEG signatures, different neurochemistry, different levels of muscle activity. If dreaming can happen in both, the neural mechanism generating it can’t simply be “whatever REM sleep does.”

One resolution comes from high-density EEG work identifying a posterior “hot zone,” a region toward the back of the brain where reduced low-frequency activity is associated with dreaming in both REM and NREM sleep [1,3]. Within that zone, spikes in high-frequency activity track specific dream content: faces, movement, spatial settings, speech [3]. Researchers have used this pattern to predict, in real time, whether someone is dreaming before waking them to ask. That’s a fairly direct piece of evidence that dreaming corresponds to something you can actually measure in the brain rather than being a purely subjective mystery.

The competing models of how dreams are generated

There are two broad camps on where dreams come from, and they don’t fully agree.

The brainstem-first view. This is the older and still influential activation-synthesis model, first proposed by Hobson and McCarley in 1977 [3]. The idea is that during REM sleep, the pons fires off essentially random bursts of neural activity, called PGO waves (named for the pons, geniculate nucleus, and occipital cortex, the structures they pass through) [3]. These waves activate the visual cortex directly, without any external sensory input to compete with, and the forebrain does its best to stitch that noise into something resembling a coherent narrative. Under this model, dreams aren’t meaningful messages. They’re the brain’s attempt to make sense of internally generated static.

The forebrain-first view. A separate line of evidence suggests dreaming can be dissociated from REM sleep altogether. Focal lesions in the forebrain have been shown to eliminate dreaming completely without changing REM sleep frequency, duration, or structure in any measurable way [3]. That points to a dopaminergic forebrain mechanism, centered on the mesocortical-mesolimbic dopamine system, as being necessary for dream generation, with the brainstem acting more as one possible trigger rather than the actual source [3]. This model, called the reward activation model, ties dreaming to the same dopamine circuitry involved in motivation and reward-seeking, which would help explain why dreams so often revolve around wanting something, chasing something, or reacting to threat.

Beyond these two, the AIM model (activation, input, modulation) tries to map consciousness across wake, NREM, and REM sleep along three variables rather than treating dreaming as an on/off switch [3]. It’s a more flexible framework, but it doesn’t settle which anatomical structures actually do the generating.

So what is dreaming actually for?

This is where things get contested, because testing a theory of dream function is hard when you can’t reliably manipulate dream content and can only rely on self-report after the fact.

Threat simulation. Proposed by Antti Revonsuo, this theory argues dreams exist to rehearse threat perception and threat avoidance, essentially a nightly flight simulator for danger [3]. The evolutionary logic: ancestral humans who mentally rehearsed responses to predators and hostile strangers had a survival edge, so the trait got selected for. Supporting evidence comes from studies showing dreams disproportionately feature unfamiliar people and negative social interactions compared to waking life, and dreams referencing genuine threats (like the COVID-19 pandemic) show more severe threatening content than dreams without those references [1]. The theory has an obvious weak spot, though: a large share of recurring dreams have no clear connection to any real threat, and a lot of dream content is simply too unrealistic to function as useful rehearsal [3].

Emotional regulation. A related but distinct idea is that dreams help process and regulate emotion, particularly negative emotion, rather than specifically rehearsing threats. Neuroimaging supports part of this: REM sleep shows increased activity in the amygdala and other limbic structures, which is thought to underlie the emotional intensity dreams often carry [1,2]. During the COVID-19 pandemic, several studies found that people with more psychological distress reported more nightmares, and that nightmare frequency tracked with worsening mental health over time [1]. That correlation runs in a useful direction for researchers: it suggests dream content isn’t random noise but actually mirrors what’s going on in someone’s waking life, which is sometimes called the continuity hypothesis.

Memory consolidation. Sleep is well established as important for memory, and PGO wave density during REM sleep has been shown to correlate with better retention after training in animal studies [3]. But this connects dreaming to a broader function of REM sleep rather than proving dreaming itself does the consolidating. It’s a subtle distinction, but an important one: something can be true about REM sleep’s role in memory without dreaming specifically being the mechanism responsible.

Unlearning. One of the stranger theories, proposed by Crick and Mitchison in 1983, argues dreams exist to erase unwanted memories rather than store useful ones [3]. Under this model, REM sleep runs a kind of reverse learning process, using PGO waves to weaken synaptic connections tied to “parasitic” memory traces that would otherwise clutter the cortex. The theory predicts that if this unlearning process breaks down, the result would resemble hallucinations or delusions, since unwanted internal activation would go unchecked [3]. It’s a difficult theory to test directly in humans, but some experimental work in mice has shown REM sleep selectively pruning certain dendritic spines while preserving others, which is at least consistent with a selective “editing” process happening during REM sleep [3].

Where the research leaves us

None of these theories fully cancels out the others, and that’s the honest state of the field. Dreams might serve multiple overlapping purposes depending on the sleep stage, brain region involved, and even the individual dreamer. What’s changed in recent years isn’t so much a settled answer as better tools: real-time EEG monitoring that can predict dreaming before someone wakes up, natural experiments like the pandemic that let researchers track dream content against measurable psychological outcomes, and improved separation of REM-specific versus NREM-specific dream features [1,2].

What research has ruled out is the older assumption that dreaming is simply a REM sleep byproduct with no independent function. Focal brain lesions can eliminate dreaming while leaving REM sleep completely intact, and vice versa, which means the two are at least partially separable processes with their own underlying circuitry [3]. Whatever dreaming turns out to be for, it isn’t just a side effect of REM sleep happening to occur.

The bigger open question, the one that shows up at the end of nearly every recent review on the topic, is still the same one philosophers and scientists have been asking for over a century: why do we dream at all? Right now, the honest answer is that nobody fully knows. But for the first time, researchers actually have the tools to start narrowing it down.

Common Questions

Do we only dream during REM sleep?

No. Dreaming is most vivid and frequent during REM sleep, but it also occurs during non-REM sleep. Late-night NREM awakenings can produce dream reports that are hard to tell apart from REM dreams, even though NREM dreams tend to be shorter and more conceptual on average.

Why can't I remember most of my dreams?

Dream recall depends on brain activity right around the moment you wake up rather than on how vivid the dream itself was. High dream recallers show stronger reactivity to nighttime sounds and stronger connectivity in memory-related brain networks immediately after waking, which suggests recall is really about how well the dream gets encoded on the way out of sleep.

Is there one accepted explanation for why we dream?

Not yet. Researchers have proposed several competing and sometimes overlapping theories, including threat simulation, emotional regulation, memory consolidation, and synaptic unlearning. Each has supporting evidence, but none has been confirmed as the full explanation, and some researchers still argue dreaming may not have an independent function at all.

Do nightmares mean something is wrong?

Not necessarily, though frequent nightmares have been linked to poorer psychological wellbeing in several studies, including research conducted during the COVID-19 pandemic. Nightmares appear to increase during periods of collective stress and have also been associated with difficulties in emotional regulation and, in some populations, increased suicidal ideation, so persistent nightmares are worth mentioning to a doctor.

References