An electroencephalogram, or EEG, is a recording of the brain’s electrical activity picked up by small sensors placed on the scalp. The technique dates back a full century: in 1924, the German psychiatrist Hans Berger recorded the first human brain electrical signal at the University Hospital in Jena and identified the rhythmic oscillations he called alpha waves.1PubMed. Hans Berger (1873-1941): the German psychiatrist who recorded the first electrical brain signal in humans 100 years ago Since then, EEG has become one of the most widely used tools in neurology and neuroscience, applied to everything from diagnosing epilepsy to guiding anesthesia, monitoring newborns, and even building brain-computer interfaces.
What the Electrodes Actually Pick Up
The brain runs on electricity. When neurons communicate, tiny electrical currents flow through and around them. Most of the signal that reaches scalp electrodes comes from the combined activity of large populations of neurons firing in rough synchrony, particularly the currents generated at synapses (the junctions between neurons). Recent biophysical modeling has shown that the scalp EEG can be thought of as a combination of synaptic currents and the electrical spikes neurons use to send long-distance signals.2PLOS Computational Biology. Contributions of action potentials to scalp EEG: Theory and biophysical simulations The voltage differences are tiny, measured in millionths of a volt, so the recording equipment needs to amplify them considerably before anyone can read the trace.
Electrodes are typically arranged on the head according to a standardized map called the International 10-20 system. The “10” and “20” refer to the percentage distances between landmark points on the skull used to position each electrode. Imaging studies using CT scans have confirmed that specific electrode positions correspond reasonably well to known brain regions underneath, though there is some variability from person to person.3Electroencephalography and Clinical Neurophysiology. Cerebral location of international 10–20 system electrode placement A routine clinical EEG might use around 20 electrodes; research setups sometimes use 64, 128, or even 256.
The Frequency Bands and What They Mean
When you look at a raw EEG, it looks like a tangle of squiggly lines. But within that tangle are rhythmic patterns at different speeds, measured in cycles per second (Hz). These are grouped into named frequency bands: delta (roughly 0.5 to 4 Hz), theta (4 to 7 Hz), alpha (8 to 12 Hz), beta (16 to 31 Hz), and gamma (above about 36 Hz).4PubMed Central. Review of electroencephalography signals approaches for mental stress assessment Each band has been linked to different mental states and cognitive processes.
Alpha waves, the ones Berger first noticed at around 10 cycles per second, tend to dominate when you are relaxed with your eyes closed.5PubMed Central. Forgotten rhythms? Revisiting the first evidence for rhythms in cognition Beta waves pick up when you are concentrating or alert. Theta waves become prominent during drowsiness and light sleep, while delta waves dominate deep sleep. Gamma oscillations, the fastest, have been associated with higher-level processing like binding together information from different senses. Berger himself was interested in how alpha and beta waves related to normal mental activity, and since then, nearly every cognitive process studied has been linked to at least one of these frequency bands.6PubMed. EEG oscillations: From correlation to causality
Epilepsy Diagnosis
The single most common clinical reason to order an EEG is epilepsy. Between seizures, the brains of people with epilepsy often produce distinctive electrical bursts called interictal epileptiform discharges, brief sharp waves that last tens to hundreds of milliseconds.7PubMed Central. Distinguishing Epileptiform Discharges From Normal Electroencephalograms Using Adaptive Fractal and Network Analysis: A Clinical Perspective Spotting these on an EEG helps confirm an epilepsy diagnosis and can give clues about where seizures originate in the brain.
Reading EEGs for epilepsy is harder than it sounds. Misidentifying normal brain activity as epileptiform discharge is a well-known cause of epilepsy misdiagnosis. A clinical validation study tested standardized criteria from the International Federation of Clinical Neurophysiology and found that requiring at least five of six specific waveform features, or using a mathematical technique called source-space analysis, could achieve about 96% specificity and over 80% sensitivity, matching the accuracy of expert neurologists.8PubMed Central. Criteria for defining interictal epileptiform discharges in EEG: A clinical validation study Patterns between seizures, including high-frequency oscillations and slow-wave activity, are also being explored as potential biomarkers that could help predict which patients are at greatest risk for future seizures.9Neurobiology of Disease. Diverse nature of interictal oscillations: EEG-based biomarkers in epilepsy
Sleep Staging
Sleep labs depend on EEG to tell the stages of sleep apart. The American Academy of Sleep Medicine recognizes distinct stages: wakefulness (W), three stages of non-rapid-eye-movement sleep (N1, N2, and N3), and rapid eye movement sleep (R). Each stage has a characteristic EEG signature. Light sleep brings theta waves and brief bursts called sleep spindles; deep sleep is marked by large, slow delta waves; REM sleep looks paradoxically similar to wakefulness on EEG, with faster, lower-voltage activity, even though the body is effectively paralyzed. The shift from one stage to another across the night follows a predictable architecture, and disruptions in that architecture can point to disorders like narcolepsy or sleep apnea.
In the Operating Room
Another major use of EEG is monitoring how deeply someone is sedated during surgery. General anesthesia suppresses brain activity in ways that show up clearly on the recording: as sedation deepens, faster frequencies give way to slower ones, and the overall pattern becomes more regular and eventually very low in amplitude. Commercial devices have been developed that translate these EEG changes into a single index number that anesthesiologists can watch in real time, aiming to keep patients in the right depth range and reduce the risk of intraoperative awareness, where a patient becomes conscious during a procedure.10PubMed Central. Monitoring the depth of anaesthesia Newer approaches are combining complexity measures and power spectrum analysis to improve accuracy and robustness across different patients and anesthetic agents.11PubMed Central. Accurate depth of anesthesia monitoring based on EEG signal complexity and frequency features
Confirming Brain Death
At the other extreme of consciousness, EEG plays a role in confirming brain death. Guidelines call for a 30-minute recording that shows complete electrocerebral silence, meaning no brain-generated activity above 2 microvolts.12PubMed. EEG guidelines in the diagnosis of brain death To make this determination reliable, the electrodes must be spaced at least 10 centimeters apart, stimulation (touch, sound) must be applied during the recording to rule out responsiveness, and clinicians must first exclude confounders like heavy sedation, severe metabolic disturbance, or hypothermia.13Annals of Clinical Neurophysiology. Electroencephalography for the diagnosis of brain death EEG alone does not determine brain death in most legal frameworks, but it is an important piece of the picture alongside clinical examination and sometimes blood-flow studies.
Event-Related Potentials in Research
Beyond clinical diagnosis, EEG is a staple of cognitive neuroscience research through a technique called event-related potentials, or ERPs. The idea is straightforward: you show a person the same stimulus many times (a tone, a flash of light, a word on a screen) and average the brain’s electrical response across all those repetitions. Random background activity cancels out, and what remains is a tiny but reliable voltage pattern time-locked to the event. ERPs index a range of sensory, cognitive, emotional, and motor processes.14PubMed Central. ERP CORE: An open resource for human event-related potential research Different components of the ERP waveform, named by their polarity and timing (like the P300, a positive peak about 300 milliseconds after a surprising stimulus), have been tied to attention, memory updating, language comprehension, and decision-making.15PubMed Central. Event-related potential: An overview
What EEG Does Not Do Well
For all its strengths, EEG has real limitations. The biggest one is spatial resolution. The skull and the layers of tissue between the brain and the scalp smear electrical signals, so pinpointing exactly where in the brain a signal originated from scalp recordings is genuinely difficult. Mathematically, this is known as the inverse problem: the number of possible combinations of brain sources that could produce the same pattern at a limited number of scalp electrodes is infinite.16PubMed Central. Electroencephalography source localization Researchers have developed methods to narrow down the possibilities by making assumptions about brain anatomy and how currents propagate.17PubMed Central. Review on solving the inverse problem in EEG source analysis These source-localization methods are useful but nowhere near the millimeter-level precision of an MRI scan.
The other constant headache is artifacts. Eye blinks, jaw clenches, scalp muscle tension, and even heartbeats all generate electrical signals that can swamp the brain activity you are trying to measure. Sophisticated mathematical techniques, like independent component analysis, can separate brain signals from these contaminants, sometimes correcting just the contaminated portions of the recording rather than throwing out entire chunks of data.18PubMed Central. Improved EOG Artifact Removal Using Wavelet Enhanced Independent Component Analysis Methods that combine different statistical approaches into a single step have also shown promise, particularly when the raw signal is noisy.19PubMed. Simultaneous ocular and muscle artifact removal from EEG data by exploiting diverse statistics Even so, artifact removal remains an active area of research, and no method is perfect.
One way around both problems is to put electrodes directly on or inside the brain. Invasive EEG, such as electrocorticography (where a grid of electrodes is placed on the brain surface during surgery), avoids the skull’s smearing effect entirely. Comparisons of invasive and non-invasive EEG recorded simultaneously from the same patients have found that invasive signal quality is 20 to over 100 times better in terms of the ratio of brain signal to artifact.20PubMed. Signal quality of simultaneously recorded invasive and non-invasive EEG Obviously, brain surgery is not a routine diagnostic option, but invasive recordings are used in presurgical evaluation for drug-resistant epilepsy and in brain-computer interface research.
AI-Assisted Seizure Detection
Reading long EEG recordings is tedious and requires trained specialists. A 24-hour epilepsy monitoring session generates an enormous amount of data, and a human reviewer can miss things. Deep learning, a branch of artificial intelligence, has been applied to automate parts of this process. Unlike traditional computer-aided detection, which relies on hand-picked signal features, deep learning models figure out the relevant features on their own from large training datasets.21PubMed Central. Epileptic Seizures Detection Using Deep Learning Techniques: A Review
Results so far have been striking in controlled settings. One deep learning framework for neonatal seizure detection achieved classification accuracy in the range of 96 to 97%, though this level of performance required large amounts of training data.22Scientific Reports. A deep learning framework for epileptic seizure detection based on neonatal EEG signals A more recent hybrid model combining convolutional neural networks with a temporal modeling architecture achieved over 98% accuracy for seizure detection on a well-known benchmark dataset and generated heatmaps showing where seizure onset occurred, information that could help guide treatment.23PubMed Central. CNN-Autoformer: Automated EEG-Based Seizure Detection and Localization Using Hybrid Deep Learning The gap between lab performance and everyday clinical use is still being bridged, but the direction is clear: AI-assisted EEG interpretation is heading toward real-world deployment.
Newborn Brain Monitoring
One of the more consequential applications of EEG is in neonatal intensive care. When a baby suffers oxygen deprivation around birth, a condition called hypoxic-ischemic encephalopathy, clinicians need to quickly assess how badly the brain has been affected and whether the infant will benefit from therapeutic cooling (lowering the body temperature slightly to protect the brain). EEG recorded in the first hours of life helps stratify these babies. A systematic review and meta-analysis found that EEG performed within the first 72 hours had good sensitivity for predicting developmental problems, making it a useful tool for identifying high-risk newborns early.24PubMed Central. Early Electroencephalography and Amplitude-Integrated Electroencephalography for the Prediction of Neurodevelopmental Outcomes in Neonates with Hypoxic Ischemic Encephalopathy: A Systematic Review and Diagnostic Test Accuracy Meta-Analysis
A simplified version of EEG called amplitude-integrated EEG (aEEG) is especially popular in neonatal units because it can run continuously at the bedside with fewer electrodes. However, its predictive power shifts depending on timing and treatment. An aEEG recorded at six hours of age has limited positive predictive value, meaning an abnormal reading at that point does not necessarily mean a bad outcome, particularly in infants who go on to receive cooling therapy.25PubMed Central. Predictive value of the amplitude integrated EEG in infants with hypoxic ischaemic encephalopathy: data from a randomised trial of therapeutic hypothermia A persistently abnormal aEEG at 48 hours or later, though, is more strongly associated with poor outcomes.26PubMed. Predictive value of amplitude-integrated EEG (aEEG) after rescue hypothermic neuroprotection for hypoxic ischemic encephalopathy: a meta-analysis The practical takeaway for families is that an early abnormal trace is a reason for close monitoring, not a definitive prognosis.
EEG in Psychiatry
Researchers have long hoped that EEG patterns might serve as biomarkers for psychiatric conditions the way blood tests work for metabolic diseases. The idea is appealing: unlike symptoms reported in a conversation, brain wave measurements are objective and quantifiable. One pattern that has received attention is frontal alpha asymmetry, a difference in alpha-wave power between the left and right frontal lobes. A study comparing patients with schizophrenia, major depressive disorder, and healthy individuals found that schizophrenia patients had significantly lower left frontal alpha activity than healthy controls, but the difference between depression patients and healthy controls was not significant.27PubMed Central. Comparison of frontal alpha asymmetry among schizophrenia patients, major depressive disorder patients, and healthy controls Findings like these are intriguing but not yet reliable enough for individual diagnosis. The variability between people is large, and no EEG marker has achieved the kind of sensitivity and specificity needed for a standalone psychiatric diagnostic test.
Neurofeedback
If EEG can measure brain activity, can you use that measurement to train people to change their own brain waves? That is the premise behind neurofeedback. During a session, your EEG is recorded in real time and displayed back to you (often as a game or visual signal), and you try to push the signal in a desired direction, say, increasing alpha power to promote relaxation or adjusting certain frequency ratios associated with attention. The underlying idea is that this learned self-regulation might produce lasting changes through neuroplasticity.28PubMed. Neurofeedback for cognitive enhancement and intervention and brain plasticity
Neurofeedback has been tried for ADHD, anxiety, depression, insomnia, and peak performance training. However, the evidence base is uneven. A comprehensive review noted that while neurofeedback is non-invasive and used as a complementary treatment for various brain dysfunctions, its validity in terms of conclusive scientific evidence has been questioned. Sessions are expensive, benefits may not be long-lasting, and improvement can take months to appear.29PubMed Central. Neurofeedback: A Comprehensive Review on System Design, Methodology and Clinical Applications Some of the positive results in the literature struggle to rule out placebo effects, since participants know they are receiving treatment. Neurofeedback is not considered a first-line treatment for any condition by major medical guidelines, though some practitioners and patients report meaningful subjective benefits.
Wearable and In-Ear Devices
Traditional EEG setups are bulky. Getting a full cap of wet-gel electrodes applied takes time and confines the person to a lab or hospital room. A growing area of development is miniaturized, wearable EEG, including devices that fit inside the ear canal. These in-ear EEG devices are unobtrusive enough to wear during normal daily activities or sleep. A comparison study found that in-ear EEG signals showed similar spectral patterns to conventional scalp recordings, with clear alpha peaks during rest and theta activity during naps. About 80% of in-ear recordings showed significant correlations with scalp EEG signals from nearby temporal electrodes.30PubMed Central. Signal quality evaluation of an in-ear EEG device in comparison to a conventional cap system Dry-contact electrode versions that require no conductive gel have also been tested, with statistically significant brain responses detected across multiple cognitive tasks, though the signals tend to be smaller in amplitude than what you get from a full scalp setup.31PubMed. Dry-Contact Electrode Ear-EEG
The appeal is enormous: imagine monitoring for seizures over days without being tethered to a hospital bed, or tracking sleep stages at home with the fidelity of a lab study. The tradeoff is signal quality. In-ear devices record from a limited patch of the brain near the temporal lobe, so they cannot capture the full spatial picture that a whole-head cap provides. Movement and muscle artifacts during daily activity also remain a challenge. Still, for specific applications where convenience matters more than full spatial coverage, the technology is rapidly closing the gap.
Privacy and the Ethics of Brain Data
As EEG moves out of the clinic and into consumer devices (headbands for meditation, gaming headsets, sleep trackers), a new set of questions emerges about who owns your brain data and what can be inferred from it. The signals collected by these devices can, in principle, be analyzed to infer information about cognitive states, emotional responses, or even personal traits that the user never intended to share. Advances in wearable technology and AI have enabled what researchers call “cognitive biometrics,” the processing and transmission of data about mental states that poses unique risks to mental privacy.32PubMed. Beyond neural data: Cognitive biometrics and mental privacy
The ethical concerns go beyond simple data security. The interaction between users and intelligent neurotechnological devices, especially closed-loop systems that adapt in real time to brain activity, raises questions about autonomy and even a person’s sense of agency.33Neuroethics. Big Brain Data: On the Responsible Use of Brain Data from Clinical and Consumer-Directed Neurotechnological Devices Legal frameworks in most countries have not caught up. Brain data is not clearly classified under existing health-data privacy laws in many jurisdictions, and current protections focusing narrowly on “neural data” may miss the broader category of cognitive biometrics collected by non-medical devices. UNESCO has begun drafting recommendations on the ethics of neurotechnology, but binding regulation remains sparse. For now, if you use a consumer EEG device, reading the data-sharing and storage terms carefully is about the only protection you have.
Recording Brain Waves in Other Species
EEG-like techniques are not limited to humans. Researchers record brain electrical activity in animals to study sleep, cognition, and neurological disease across species. Work in birds has produced some fascinating comparisons with mammalian brains. Despite having a fundamentally different brain architecture (bird brains are organized in clusters of neurons called nuclei rather than the layered cortex mammals have), electrical slow waves during non-REM sleep propagate through the avian brain in complex three-dimensional patterns.34PubMed Central. Plumes of neuronal activity propagate in three dimensions through the nuclear avian brain Some aspects of avian sleep look strikingly similar to mammalian sleep, like the propagation of slow waves in regions receiving input from the thalamus. Others differ: for instance, the characteristic sleep spindles seen in mammalian EEG recordings were not detected in the visual regions of the avian brain.35Sleep. Intra-“cortical” activity during avian non-REM and REM sleep: variant and invariant traits between birds and mammals These comparisons help researchers understand which features of brain electrical activity are ancient, shared across vertebrate evolution, and which are more recent innovations specific to certain lineages.