EEG Biofeedback

Alpha discrimination experiments

Joseph Kamiya's work on the alpha rhythm is the usual starting point for EEG neurofeedback. A 2012 replication study in Conscious and Cognition cites his conference paper of 1962 on conditioned discrimination of the EEG alpha rhythm, his 1968 article in Psychology Today and a 1969 chapter on operant control of alpha in a volume on altered states of consciousness. The replication authors describe these reports as the first demonstration that people could learn to tell alpha activity from other EEG states.

In Kamiya's design, a person was prompted to say whether a short EEG episode was alpha or non-alpha and received feedback on each answer. The replication cites Kamiya's finding that nine of 12 subjects reached the criterion within seven sessions. Its own study found that 40 of 106 participants (38%) reached criterion, and that 75% of those who completed nine or more sessions did. The authors note at least three failed attempts by other researchers to replicate the effect before theirs.

Sensorimotor rhythm and seizure research

Barry Sterman's early animal work trained a 12 to 15 Hz rhythm over the sensorimotor cortex, which he called the sensorimotor rhythm (SMR). In a 2010 review in the Cleveland Clinic Journal of Medicine, Sterman writes that the SMR was linked to motor inhibition and to sleep spindle activity, and that an anticonvulsant effect on drug-induced seizures in cats and monkeys was found by accident during these studies. He then describes human epilepsy studies that documented fewer seizures and EEG pattern normalization.

Sterman's human epilepsy work is often dated to a 1974 paper in Epilepsia by Sterman, MacDonald and Stone. That paper was not opened for this record. It appears in the reference list of a 2009 ADHD meta-analysis by Arns and colleagues, which is cited here only for that reference.

Clinical use and reporting standards

The 2009 ADHD meta-analysis by Arns and colleagues states that neurofeedback treatment of ADHD was first reported in 1976. Its abstract adds that many practitioners used the technique at that time and that the level of evidence was still unclear.

A 2009 meta-analysis of epilepsy studies by Tan, Thornby and Hammond indexed 63 EEG biofeedback reports from 1970 to 2005 and included ten with enough outcome information to pool. Nine of the ten reinforced the sensorimotor rhythm and one trained slow cortical potentials. The authors describe most of the 63 reports as case series with too few subjects to establish efficacy.

In 2020 a group of researchers that included Sterman published the CRED-nf checklist, a consensus on the reporting and experimental design of clinical and cognitive-behavioural neurofeedback studies, in the journal Brain. It is a set of reporting and design standards for studies, not a finding about whether the method works.

Names used for the method

Researchers and practitioners use several names for the same broad approach. Tan and colleagues write that EEG biofeedback, or neurofeedback, normalizes or enhances EEG activity by means of operant conditioning. Sterman's 2010 review uses EEG operant conditioning and biofeedback as parallel terms. The shorter term neurofeedback appears in the title of the 2016 ADHD meta-analysis by Cortese and colleagues.

The word biofeedback places the method within a wider group of techniques that show a person a body signal so that the signal can be changed. Sterman's review traces EEG operant conditioning, or biofeedback, from animal research to its use in epilepsy. In this record, EEG biofeedback and neurofeedback refer to the brain-signal variant, not to muscle, skin or heart-rate training.

Protocol and target terms

The target signal also varies by study. The 2016 ADHD meta-analysis describes frequency-band training as the most common neurofeedback approach among its included trials. The epilepsy meta-analysis includes one study that trained slow cortical potentials rather than the sensorimotor rhythm, so findings for one target cannot be assumed for the other.

Alpha state discrimination is a separate research tradition. Frederick and colleagues describe it as training awareness of the subjective correlates of EEG states, in contrast to much research since Kamiya's work, which has focused on training voluntary control of EEG measures.

Observable features of a session

In EEG neurofeedback, sensors on the scalp record electrical activity, and the signal is processed and returned to the participant as feedback. Studies describe this feedback as a continuous or threshold-based signal, and the choice between these methods has been debated in the research literature. Training is repeated across multiple sessions, and the number of sessions and total training time vary between studies.

The 2025 meta-analysis of randomized trials in children with ADHD grouped trials by total training time and reports that training beyond about 1,260 minutes was associated with larger effects on inhibitory control and working memory. Protocols differ in the signal trained, the target band or state, the threshold for feedback and the number of sessions, and a study report should state these choices.

Controls used in studies

Controlled trials compare neurofeedback with a sham or with an active comparison. The CRED-nf checklist lists placebo-type controls, such as sham neurofeedback or feedback from an unrelated brain signal, and active non-neurofeedback controls, such as similar computerized training. The 2016 ADHD meta-analysis found that effects were not significant in trials with active or sham controls.

ADHD symptoms and executive function

Practitioners and some researchers describe neurofeedback as a way to reduce inattention, impulsivity and hyperactivity in ADHD. The 2016 meta-analysis of 13 randomized trials found small positive effects on ADHD symptoms only when outcomes were rated by assessors closest to the treatment setting. Effects were not significant when probably blinded raters were used, and laboratory measures of inhibition and attention were not significant.

Improvement in executive function is a second claimed use. Based on the abstract read for this record, a 2025 meta-analysis of 17 randomized trials with 939 children reports improvements in inhibitory control and working memory, with a more sustained effect on working memory. The same analysis is an abstract-level summary of the included trials, and the individual trials were not reviewed here.

Seizure frequency in epilepsy

Epilepsy is the other use with a substantial clinical literature. The meta-analysis by Tan and colleagues reports a significant overall reduction in seizure frequency, with a pooled effect of -0.233 and 64 of 87 patients reporting fewer weekly seizures. All the included patients had seizures that were not controlled by medical therapy.

Sterman's 2010 review describes significant reductions in seizure incidence and severity in human subjects, with EEG pattern normalization, and presents these as the basis for clinical application. That is the author's own summary of his program and of other studies, and this record did not verify its individual results.

Other reported uses

Practitioners also describe neurofeedback for anxiety, depression, sleep, performance and other concerns. The sources opened for this record do not establish outcomes for those uses, and no finding for them is attributed to this method here.

ADHD trials and meta-analyses

Cortese and colleagues searched databases through August 2015 and included 13 randomized controlled trials with 520 participants with ADHD. They used random-effects models and the Cochrane risk-of-bias tool, and they found the risk of bias unclear for many domains in most studies. Only four studies directly assessed whether learning occurred after training. The authors conclude that evidence from well-controlled trials with probably blinded outcomes does not currently support neurofeedback as an effective ADHD treatment.

An earlier meta-analysis by Arns and colleagues in 2009 pooled ADHD studies and reported large effect sizes for prospective controlled and pre-post designs, and a medium effect for hyperactivity. Randomized studies in the same analysis showed a lower effect for hyperactivity, which the authors interpret as suggesting that hyperactivity is sensitive to nonspecific treatment factors. The difference between design types is the reason design comparisons matter when reading these pooled results.

The 2025 meta-analysis of executive function reports 17 randomized trials with 939 children. It reports a significant gain in inhibitory control and working memory, but its p value for global executive function is 0.055, which is above the conventional 0.05 threshold. Its clearer findings are therefore for the two specific domains.

Epilepsy studies

Tan and colleagues found 63 reports from 1970 to 2005 and pooled ten that had enough outcome information. The pooled result was a significant reduction in seizure frequency under a random-effects model. Because the included patients had seizures not controlled by medication, the result cannot be applied to people whose seizures are already controlled, and the authors note that the studies were small.

Sterman's 2010 review summarizes later group and single-case studies as well-controlled, but it is a single-author summary and does not present a new pooled analysis. The individual trials it describes were not checked for this record, so its description of their quality is recorded as the author's view.

Reporting quality and remaining questions

The CRED-nf checklist was developed to improve the design and reporting of neurofeedback studies, including control conditions. It does not state whether the method works. Its existence reflects the concern in the reviews that many earlier studies were small, uncontrolled or hard to compare.

The sources reviewed here do not cover neurofeedback for anxiety, depression, sleep or performance in healthy people in any depth, and this record does not present those uses as evidence-supported. Trials that use blinded outcome measures and active comparators are the main gap the reviews identify.

Harms reported in the sources reviewed

The abstracts and sections opened for this record do not report adverse-event rates in a form this record can summarize. The reviews describe outcomes, risk of bias and controls, but they do not give harm data in the parts read here. This record therefore does not state how often side effects occur or how serious they are.

Cortese and colleagues found that symptom effects were not significant when raters were probably blinded. Reports of benefit based on observers who knew the treatment condition should be read with that limit in mind, and expectation can shape what is recorded.

Epilepsy and medication

The epilepsy studies reviewed by Tan and colleagues enrolled patients whose seizures were not controlled by medical treatment. That context does not support changing, stopping or delaying antiseizure medication or specialist epilepsy care, and this page does not give advice on those decisions. Neurofeedback is not a substitute for medical assessment or a prescribed treatment.

Seizures, sudden changes in awareness, new neurological symptoms or a marked change in a child's attention or behavior need prompt assessment by a qualified clinician. Neurofeedback should not be used in place of that assessment, and it does not replace treatment prescribed for ADHD or epilepsy.

Questions to ask a provider

A person considering neurofeedback can ask which signal is trained, which control or comparison is used, how many sessions are planned, how progress will be measured, and what training the provider has completed. Professional training and certification arrangements are set by professional bodies and are separate from research evidence. This record did not verify a professional register or training standard for the method.