Brain Waves and EEG

What EEG actually records

Start with the physics, because almost every misconception downstream comes from getting this wrong. When Hans Berger recorded the first human EEG in the 1920s, he was picking up voltage differences between electrodes on the scalp. Those voltages are produced by electrical currents flowing through brain tissue. The question is which currents.

The answer is: mainly the postsynaptic potentials of cortical pyramidal neurons. When a synapse on a pyramidal cell's dendrite is activated, current flows into or out of the dendrite, and a compensating current flows in the opposite direction elsewhere along the cell. The cell behaves, electrically, like a small dipole. Crucially, pyramidal neurons in the cortex are arranged in parallel, all with their long apical dendrites pointing the same way, perpendicular to the cortical surface. Because they are aligned, their tiny individual dipoles add up rather than cancelling out. Millions of them, active together, produce a field large enough to be measured through the skull.

Note what EEG does not record. It does not directly capture action potentials: spikes are too brief and too asynchronous to sum effectively at the scalp. It is strongly biased toward the cortical surface, especially the gyral crowns, and it is largely blind to deep structures such as the hippocampus and basal ganglia in ordinary scalp recording. When popular articles speak of "hippocampal theta" measured with a consumer headband, they are describing something that headband cannot see.

Strengths and hard limits

EEG occupies a specific and valuable niche among brain measurement methods, and understanding it means understanding a trade-off.

The temporal precision is genuinely outstanding. Where fMRI measures a haemodynamic response that lags neural activity by several seconds and blurs it across several more, EEG follows neural events millisecond by millisecond. If your question is when, EEG is often the best tool available.

The spatial limitation is equally real and cannot be engineered away by adding electrodes. The skull is a poor conductor and acts as a spatial low-pass filter, spreading each source across a wide patch of scalp. Worse, the inverse problem, working backwards from a scalp voltage pattern to the brain sources that produced it, has no unique solution: infinitely many source configurations can produce the same surface map. Source localisation algorithms exist and are useful, but they work by imposing assumptions, not by solving the problem outright. Compare with brain imaging methods, where fMRI makes the opposite trade.

The frequency bands, honestly

EEG oscillations are conventionally sorted into frequency bands. These names are useful shorthand and they are used by every clinical and research laboratory. But they are descriptive conventions, boundaries drawn on a continuum for convenience, and they do not carve nature at joints as cleanly as the popular literature pretends. Here is what each band actually correlates with, with the qualifications intact.

Two warnings about band talk. First, the boundaries vary between laboratories: one paper's alpha upper limit is another's lower beta. Second, and more importantly, the mapping from band to mental state is many-to-many, not one-to-one. Theta appears in drowsiness and in effortful memory encoding, two states that could hardly be more different. Alpha appears in relaxed idling and in effortful inhibition of distractors. If you are told that a given frequency is a given mental state, you are being sold something.

Event-related potentials

Band analysis is not the only, or even the most powerful, use of EEG. The event-related potential is a different and in many ways more rigorous technique. The idea is simple. Present the same kind of stimulus many times, record the EEG each time, and average the segments together. Activity unrelated to the stimulus is random with respect to its timing and averages toward zero; activity that is reliably time-locked to the stimulus survives. What is left is a sequence of positive and negative deflections, each reflecting a stage of processing.

1. Early sensory components

   Deflections in the first hundred milliseconds or so, such as the visual P1 and N1, index basic sensory registration and are strongly modulated by attention. They show that attention operates early, at the level of perception, not only at the level of decision.

2. The P300

   A large positive deflection peaking roughly 300 milliseconds or later after a rare or task-relevant stimulus, discovered in the 1960s. Its amplitude tracks the surprise and significance of the event and its latency tracks the time needed to categorise it. It is one of the most-studied signals in cognitive neuroscience and is the basis of some brain-computer interface spellers.

3. The N400

   A negative deflection around 400 milliseconds that is larger when a word is semantically incongruent with its context: "I take my coffee with cream and socks" produces a big N400. It is a workhorse of the neuroscience of language and demonstrates that meaning is being evaluated with millisecond precision.

4. The error-related negativity

   A sharp negative deflection appearing within about a hundred milliseconds of making an error, generated in or near the anterior cingulate cortex, and often occurring before the person consciously realises they have erred. It is a direct electrophysiological signature of performance monitoring.

Clinical uses

EEG earns its place in medicine on two grounds above all.

Epilepsy

This is the flagship application. Seizures are, at bottom, episodes of pathologically excessive and hypersynchronous neuronal firing, and hypersynchrony is precisely what EEG is best at detecting. Interictal spikes and sharp waves between seizures help establish a diagnosis and localise the focus; capturing an actual seizure on video-EEG is the reference standard for classifying seizure type; and in surgical candidates, intracranial EEG maps the tissue that must be removed. No other non-invasive method comes close.

Sleep staging

Polysomnography, the standard sleep study, is built around EEG, supplemented by eye movement and muscle tone recordings. Stages are scored from the EEG signature: spindles and K-complexes define N2, delta dominance defines N3, and a wake-like trace combined with atonia and eye movements defines REM. See sleep and the brain.

Other established uses

EEG contributes to the assessment of encephalopathy, to intraoperative and intensive-care monitoring including the detection of non-convulsive status epilepticus, and to the ancillary assessment of severe brain injury. What it is not is a general-purpose window into thought, and it is not a diagnostic test for psychiatric or developmental conditions. Claims that a scalp EEG can diagnose ADHD, depression or autism as a standalone test are not supported by the evidence base.

Where the pseudoscience begins

The commercial exploitation of EEG follows a consistent recipe. Take a real phenomenon (cortical rhythms exist), attach a false premise (a band equals a mental state), add an unjustified causal claim (drive the band and you produce the state), and sell a device or an audio file. Each step is wrong in a way that is easy to miss if you do not know the underlying science, which is exactly why the recipe works commercially.

Binaural beats. The physical phenomenon is genuine: present 200 Hz to one ear and 210 Hz to the other, and the auditory system generates a perceptual beat at 10 Hz. The leap that follows is not genuine. The claim that this beat entrains cortical oscillations into the alpha band, and that alpha entrainment then produces relaxation, focus, or improved cognition, is weakly supported. Studies report inconsistent and generally small effects, methodological quality varies widely, effects are frequently not distinguishable from expectation and relaxing-music effects, and there is no credible evidence for cognitive enhancement. The honest summary is: interesting auditory phenomenon, no established cognitive benefit.

"Brainwave entrainment" and isochronic tones. The same logic, with the same problems. Rhythmic sensory stimulation can produce a steady-state evoked response at the driving frequency; this is a well-known and real effect, and it is used deliberately in research. But a steady-state response in sensory cortex is not the same thing as putting your brain "into an alpha state", and it does not follow that any cognitive or emotional benefit results.

Consumer "alpha training to raise IQ". This is the least defensible of the family. It requires all of the above errors plus one more: that a band, if increased, would increase general intelligence. There is no good evidence that neurofeedback training of alpha or any other band raises IQ. The wider neurofeedback literature is heavily affected by weak controls, and where properly blinded sham-controlled designs have been used, benefits have frequently failed to separate from placebo. Neurofeedback has some legitimate research standing in specific clinical contexts, but "train alpha, get smarter" is not one of them.

Consumer headsets that "read your mind". Single-electrode or few-electrode headbands, worn over hair, in a moving body, in an electrically noisy room, pick up an enormous amount of muscle artefact, eye movement artefact and mains interference. A jaw clench produces a huge signal. Some of these devices are legitimately useful as relaxation trainers or as toys. None of them are reading your thoughts, and the "focus score" they display is typically a proprietary and unvalidated composite.

Myths about brain waves

Related topics