Neurofeedback and Biofeedback
Linas JuozenasShare
Train the signal—then prove that life changed
Neurofeedback and biofeedback make hidden physiology visible. A person sees or hears a live estimate of brain activity, breathing, muscle tension, skin conductance or heart rhythm and practises changing it. That loop can teach real self-regulation. But a moving graph is not yet evidence that attention, anxiety, sleep, rehabilitation or intelligence improved.
Feedback, not stimulation
Biofeedback is the broad family: a physiological measure is translated into timely information that helps a person learn voluntary regulation. Neurofeedback is the subset using a neural measure, commonly scalp EEG and sometimes real-time fMRI or fNIRS. The person is trained; the device does not directly add intelligence or insert a mental state.
This differs from neuromodulation such as electrical or magnetic stimulation, which applies energy to nervous tissue, and from passive monitoring, which may display a score without creating a structured learning loop.
Target engagement is not treatment success
A person may learn to increase an EEG feature yet show no meaningful change in ADHD symptoms. Another may feel calmer after sham feedback because the ritual, expectation, breathing, coaching and scheduled rest were helpful. Both observations matter, but they answer different questions.
The strongest studies therefore test the signal, the learning process and the real-world outcome separately, using credible comparison conditions and assessors who do not know which training was received.12
A closed loop contains more than a sensor
Neurofeedback is best understood as a learning experiment. A physiological event must be measured, converted into feedback quickly enough to be useful, linked contingently to the person’s own activity and practised until regulation can survive outside the apparatus.
Operant learning—with human interpretation
When the display changes reliably after a desired physiological pattern, that change can act as reinforcement. Over repeated trials, useful strategies may become easier to reproduce. Neurofeedback research also implicates attention, motivation, interoception, reward learning, metacognition and explicit strategy formation; “the brain rewards itself” is too simple.1
Learning is not guaranteed. A 2025 meta-analysis of 55 participant groups found that acquisition varied with the measurement method, feedback complexity, rehearsal and targeted EEG feature. The field calls persistent failure to modulate the target the neurofeedback inefficacy problem—a reminder that non-response may reflect the protocol, signal, instruction, person or analysis rather than deficient effort.3
Four outcomes that should be reported separately
These outcomes can dissociate. Successful signal regulation without symptom change is scientifically informative. Symptom improvement without target learning may point to nonspecific benefits. A responsible report does not erase either result.
Different signals answer different questions
“Biofeedback” does not name one treatment. Each modality observes a different part of physiology, with its own spatial resolution, timing, artifacts, costs and plausible uses.
| Modality | Feedback feature | Useful strength | Central limitation |
|---|---|---|---|
| EEG neurofeedback | Voltage patterns at the scalp, often summarized as band power, slow cortical potentials or connectivity estimates. | Millisecond timing; comparatively portable and affordable. | Signals mix many neural sources and are easily contaminated by eyes, jaw, forehead, movement and reference choices. |
| Real-time fMRI | Blood-oxygen-level-dependent activity in a region or network. | Whole-brain coverage and finer spatial localization than scalp EEG. | BOLD is an indirect, delayed metabolic signal; scanners are expensive, noisy and motion-sensitive.4 |
| fNIRS | Changes in oxygenated and deoxygenated haemoglobin near the cortical surface. | More mobile and tolerant of some movement than fMRI. | Limited depth and coverage; scalp blood flow and motion can distort estimates. |
| HRV biofeedback | Beat-to-beat timing, breathing and derived heart-rate variability measures. | Portable window into cardiorespiratory regulation and baroreflex training. | HRV depends on breathing, posture, age, fitness, illness, medication, recording length and algorithm; one score is not a diagnosis. |
| EMG biofeedback | Electrical activity produced by contracting muscles. | Directly useful for muscle awareness, relaxation and some rehabilitation tasks. | Reduced tension is not automatically reduced pain or restored function. |
| EDA / temperature | Skin conductance or peripheral temperature associated with autonomic arousal. | Simple, responsive feedback for learning awareness of arousal. | Changes are nonspecific and strongly shaped by room conditions, movement and individual physiology. |
Frequency bands are summaries—not substances
EEG software often displays delta, theta, alpha, beta and gamma power. These are conventional frequency ranges applied to a continuous, mixed signal. Alpha can vary with eye closure and sensory attention; theta can vary with drowsiness, memory and control; high-frequency activity is especially vulnerable to muscle artifact. A theta-to-beta ratio cannot by itself diagnose ADHD or identify intelligence.
Modern spectral methods also distinguish oscillatory peaks from the aperiodic background of the signal. If that background changes, a band-power ratio can change even when the presumed rhythm has not changed in the expected way.5
Latency matters, but so does truthfulness
Feedback should arrive soon enough for the learner to connect it with an internal action. Yet very fast feedback built from a noisy or poorly defined feature teaches the person to control an artifact. Good systems disclose filtering, delay, artifact rejection, missing data and how rewards are calculated.
What would convince us that the loop caused the benefit?
Neurofeedback is unusually vulnerable to plausible alternative explanations because it combines technology, repeated coaching, expectation, practice, attention and often an impressive visual interface.
Controls must match the question
A waitlist asks whether the complete programme is better than waiting. An active control can match time, coaching and engagement. A sham control may replay another person’s signal, use delayed feedback or make reward independent of the current target. Each has weaknesses: sham may be detected; a poorly designed active control may be less credible; a “placebo” can still contain helpful breathing or attention practice.
Blinded ratings matter when expectations are strong. Objective tasks can help, but they also have practice effects and may not represent everyday functioning. The CRED-nf consensus therefore recommends prespecifying the feedback signal and control condition, testing regulation success, reporting adverse effects and separating experimental, behavioural and clinical outcomes.2
- DefineOne observable target and one meaningful outcome
- BaselineRepeated measurement before training
- ControlCredible sham or active comparison
- BlindOutcome assessors where possible
- TransferTest without feedback
- Follow upAsk whether benefit lasts
The dashboard is part of the intervention, not the verdict.
ADHD: real training, uncertain treatment advantage
EEG neurofeedback for ADHD has been studied for decades, often using theta/beta, sensorimotor-rhythm or slow-cortical-potential protocols. Earlier positive reviews and unblinded parent ratings helped build enthusiasm. Stronger controls have produced a more restrained conclusion.
The 2025 synthesis
A systematic review and meta-analysis of 38 randomized trials involving 2,472 participants found no meaningful group-level benefit on probably blinded ADHD ratings or neuropsychological outcomes overall. Restricting analyses to established standard protocols produced a small symptom effect, and processing-speed outcomes showed a small signal, but the authors concluded that blinded evidence does not support neurofeedback as a stand-alone ADHD treatment.6
A rigorous sham trial
In a large double-blind trial with 13-month follow-up, both active theta/beta neurofeedback and sham training were associated with improvement, but deliberate regulation of the targeted ratio did not produce a specific clinical advantage. This is exactly why “participants improved” cannot answer whether contingent EEG feedback caused the improvement.7
Personalization is being tested
A 2026 double-blind trial randomized 48 children to personalized upper-alpha training or sham feedback. Neurofeedback produced robust EEG learning and selective gains in motor speed and reaction time, yet parent-rated core ADHD symptoms did not improve more than sham. The result is promising for studying target engagement while again arguing against neurofeedback alone as an adequate treatment.8
What this means for families and adults
- Do not replace an effective medication, behavioural intervention, educational support or clinical plan on the basis of an EEG map or testimonial.
- Ask whether the goal is symptom relief, a specific cognitive process, self-awareness, medication reduction or something else—and measure that outcome directly.
- Expect providers to explain the exact protocol, comparison evidence, likely time and cost, non-response criteria and how benefit will be distinguished from repeated attention.
- Treat “personalized” as a hypothesis requiring validation, not as proof that any qEEG-selected target is uniquely correct.
Some people may still value the structured practice, agency or attention to internal state. That value should be described honestly rather than converted into a universal disease-treatment claim.
HRV biofeedback: a more direct skill, not a universal stress meter
Heart-rate variability describes variation in the time between successive heartbeats. During slow breathing, heart rate normally rises and falls with respiration. Training can amplify and stabilize this oscillation, often near the frequency at which breathing and the baroreflex interact most strongly.
What is trained
A chest strap, ECG or optical sensor estimates beat intervals while a breathing guide and HRV display provide feedback. The learner experiments with gentle pace, depth and ease rather than trying to force the highest possible score.
Why it may help
Repeated paced breathing may strengthen awareness of arousal, improve cardiorespiratory coordination and provide a practised route from activation toward steadier regulation. The mechanism is more specific than “activating calm.”9
What a score cannot prove
Higher HRV during one session does not certify emotional health, resilience or safety. Comparisons across people are easily distorted by age, fitness, posture, breathing, medication, illness, sensor quality and proprietary calculations.
Evidence is encouraging, but not uniform
A 2017 meta-analysis found large average reductions in self-reported stress and anxiety, but included only 24 studies and 484 participants and called for better controlled research.10 A broader 2020 review also reported benefits across emotional, physical and performance outcomes, while emphasizing considerable variation in protocols and applications.11
A newer review of remote HRV biofeedback included 18 studies and 1,352 participants. It found improvement in depressive symptoms and HRV, but no significant pooled effect on perceived stress; anxiety evidence came from only three studies, two nonrandomized and at high risk of bias. Only one included study was rated low risk of bias.12 This does not cancel earlier findings—it shows why population, comparator, protocol and outcome must be named.
“Resonance frequency” is useful, not sacred
Many protocols estimate an individual breathing rate that maximizes heart-rate oscillation, commonly somewhere around six breaths per minute. A systematic review of 143 studies found three major approaches—individual resonance assessment, other individualized methods and preset slow breathing—and also found that nearly two-thirds of studies reported methods too incompletely for confident replication.13
In a 2026 four-week randomized comparison, individualized resonance-frequency breathing and fixed 0.1-Hz breathing produced similar reductions in self-reported stress, anxiety and depression; neither clearly outperformed the other, and resting HRV did not change. Individual calibration may still matter for some aims, but a more elaborate setup is not automatically better.14
Promising signals across conditions—different certainty in each
A protocol useful in stroke rehabilitation cannot validate an anxiety headset. Evidence must remain attached to the exact modality, target, population, comparator and outcome.
| Area | What has been studied | Current reading | Main caution |
|---|---|---|---|
| PTSD | EEG and real-time fMRI neurofeedback targeting arousal or emotion-related networks. | Very low to low certainty | A 2025 review found moderate-to-large EEG effects against passive controls, but sham-controlled fMRI trials showed no benefit and certainty was downgraded for bias, imprecision and conflicts.15 |
| Depression | Real-time fMRI feedback from emotion-related regions and networks, often added to care. | Promising adjunct | Small heterogeneous studies suggest people can sometimes regulate targets and symptoms may improve; protocols, controls and access remain inconsistent.16 |
| Insula regulation | Real-time fMRI training of a region involved in interoception, salience, emotion and pain. | Target learning possible | A 2025 review of 25 studies found evidence of learned regulation and some transfer, but a trainable region is not one universal treatment mechanism.17 |
| Insomnia | Sensorimotor-rhythm EEG training compared with placebo feedback. | No specific advantage shown | In a double-blind study, both real and placebo feedback improved subjective sleep, while neurofeedback did not produce specific objective sleep benefits.18 |
| Stroke rehabilitation | Combined EEG–fMRI feedback during motor imagery in selected chronic stroke survivors. | Early clinical promise | A 2025 randomized trial had 30 completers and suggested motor benefits, but it was small, specialized and limited to people meeting neurophysiological criteria.19 |
Why an inactive comparison can exaggerate confidence
Feedback programmes supply scheduled contact, a coherent explanation, repeated practice and a sense of agency. Those ingredients may be genuinely helpful. Yet an effect against waiting does not reveal whether the displayed brain signal was necessary. Sham and active controls become especially important when the intervention is expensive, makes disease claims or asks patients to delay established care.
Clinical translation requires more than a significant average
Readers should look for absolute change, responder definitions, everyday function, adverse effects, durability and the range of individual outcomes. A statistically significant group difference can be too small to matter; a null average can conceal a subgroup worth studying. Subgroup claims, however, should be tested prospectively rather than invented after results are known.
Regulating a state is not manufacturing intelligence
Neurofeedback may help some people practise a particular attentional, arousal or sensorimotor state. That is a narrower claim than improving general intelligence, wisdom, creativity or expertise.
Near transfer
Training a signal may improve closely related regulation or a task repeatedly practised during the programme. This is valuable when the target has a clear functional link, but it may partly reflect task familiarity.
Far transfer
Claims about broad intelligence, grades, leadership or creativity require improvement on unfamiliar measures and real activities after feedback is removed. That evidence is much harder to establish.
Consumer mindfulness neurofeedback illustrates the gap. A 2025 meta-analysis found a modest reduction in psychological distress, possibly placebo-related, but no conclusive improvement in cognition, mindfulness, physiological health or the proposed neural targets. Nineteen of 21 included studies did not assess adverse events.20
This does not make every consumer device useless. A headset can supply a timer, ritual, reminder or responsive sound that encourages practice. The honest description is “technology supporting a routine,” not “direct control of the brain” unless target engagement and added benefit have been demonstrated.
Protect ability from misuse of measurement
Neurofeedback data should never be used to rank a person’s intellectual worth, infer honesty or punish unconventional attention patterns. Quietness, autism, motor limitations, anxiety and atypical communication can change how someone performs or produces artifacts without diminishing intelligence. Every person deserves dignity; exceptional knowledge and original judgement additionally deserve safety, autonomy, health, uninterrupted thought, fair credit and supportive collaboration.
Technology is most humane when it helps a mind express itself—not when a score is allowed to define the mind.
A careful ten-minute HRV practice
The most defensible home experiment is simple cardiorespiratory biofeedback—not self-treatment of a neurological or psychiatric disorder. The aim is to discover whether gentle paced breathing improves regulation and daily function, not to maximize a proprietary “coherence” badge.
Use comfort as a constraint
A starting pace near five to seven breaths per minute is common in research, but it is not a universal prescription. Breathing should remain smooth and modest. Light-headedness, tingling, air hunger, chest discomfort or rising panic are reasons to return to ordinary breathing and stop the session. People with significant cardiac, respiratory, fainting, pregnancy-related or other medical concerns should discuss paced breathing with a qualified clinician.
Measure what matters after the display closes
- Could you reproduce the steadier breathing without visual guidance?
- Did recovery after an ordinary stressor become faster or easier?
- Did sleep, concentration or anxiety improve consistently—not just during the attractive exercise?
- Did the practice create dependence on checking a score, frustration or avoidance?
A four-week personal test
Week 1: establish baseline variation without changing routines. Week 2: practise five to ten minutes on four days. Week 3: use the skill once in a real low-risk situation without the device. Week 4: compare repeated outcomes and decide whether to keep, modify or discontinue it.
Change one main variable at a time. Do not start neurofeedback, a new supplement, sleep restriction and medication changes together and then credit whichever explanation feels most exciting.
How to evaluate a clinic, course or consumer device
A professional title, colourful qEEG image or expensive headset does not by itself establish competence or efficacy. Good providers make their reasoning inspectable and make it easy to stop.
The SIGNAL checklist
Questions for a provider
- Is the goal clinical treatment, rehabilitation, performance support or general wellness?
- Which published protocol or rationale supports the target?
- How will regulation learning and symptom change be measured separately?
- What happens if the person cannot learn the target?
- Who reviews medications, diagnoses and adverse changes?
- Are financial interests, device partnerships and unpublished methods disclosed?
Questions for a device
- Does it provide raw or interpretable data, or only a hidden score?
- How does it identify eye, jaw, movement, poor contact and missing beats?
- Was the exact model and software version validated against an appropriate reference?
- Does a firmware update change the score without revalidation?
- Can sessions run locally, and can cloud storage be refused?
- Can data be exported and permanently deleted without losing unrelated features?
Be cautious with universal brain maps
Group averages do not define an ideal brain for every individual. A deviation from a reference database may reflect age, alertness, medication, sleep, eye state, recording choices or ordinary variation. A qEEG can be a measurement tool; it should not be used as a stand-alone diagnosis, personality test or deterministic explanation for every symptom.
Likewise, “FDA registered,” “medical grade,” “clinically tested” and “uses an FDA-cleared sensor” are not interchangeable with clearance or approval of the exact device for the exact claim. Ask for the regulatory product code, intended use and decision record.
Regulation, neural privacy and freedom from covert control
A feedback system can collect time-stamped physiology, behavioural responses, app activity and inferred states. Even when a consumer EEG cannot read a sentence in the mind, repeated neural and biometric data can still be sensitive and revealing when combined with identity, location, sleep or work records.
Medical-device status follows intended use
United States regulation defines a biofeedback device as an instrument providing a visual or auditory signal corresponding to physiological parameters so that a patient can learn voluntary control. The narrow exemption from premarket notification applies under specified conditions to prescription, battery-powered devices indicated for relaxation training and muscle re-education; broader diagnosis or treatment claims can change the regulatory pathway.21
FDA’s current classification lists biofeedback devices under Class II and recognizes IEEE 2010-2023, a technical recommended practice for EEG neurofeedback systems. Recognition of a standard can support device quality; it does not prove that a particular protocol treats ADHD, anxiety or another disorder.22
In the European Union, whether software or hardware is a medical device depends on its intended medical purpose under the Medical Device Regulation. Data processing must also comply with the General Data Protection Regulation when personal data are involved, including health or biometric data under applicable conditions.2324
Consent must cover the closed loop
A person should know not only that data are recorded, but what feature is inferred, how the interface changes in response, whether an AI model learns from sessions, who can view the results, how long data persist and whether the system is designed to influence behaviour. Consent to wear headphones is not consent to hidden mental-state profiling.
A modern neural-data standard
UNESCO’s 2025 Recommendation on the Ethics of Neurotechnology extends concern beyond direct neural recordings to indirect data that can support mental-state inference. It calls for prior, free and informed consent, data minimization, purpose limitation, security and the ability to access, correct, erase or suspend processing. It also rejects coercive or manipulative use and gives special attention to workplaces and children.25
The OECD similarly recommends clear information about collection and use, meaningful choices over sharing, access and deletion, protection from discrimination and avoidance of hype. Brain-related data should not become a hidden condition of employment, insurance, education or ordinary participation.26
Private by default
Collect the minimum signal needed, prefer local processing where practical, encrypt transfers and storage, separate identity from research data and set short retention periods.
No engineered compliance
Feedback should support a chosen goal. Employers, schools, partners or therapists should not secretly tune interfaces to make a person more obedient, disclose more or suppress legitimate disagreement.
Human-readable control
Users need plain explanations, visible recording indicators, pause and deletion controls, an exportable record and a way to use core functions without surrendering unrelated data.
Safety, adverse effects and myths worth retiring
Non-invasive feedback is often described as harmless because it does not inject a drug or apply current. Yet training can still change breathing, arousal, attention, expectations and behaviour, while electrodes and algorithms can produce misleading interpretations.
Monitor the whole person, not only the target
Headache, fatigue, frustration, dizziness during breathing practice, sleep disruption, agitation, increased anxiety or other unwanted changes should be recorded rather than rebranded as proof that the brain is “reorganizing.” Adverse-event reporting has often been weak: the consumer mindfulness review found that most included studies did not assess it at all.20
Clinical programmes should define who supervises emerging symptoms, when training pauses and how existing care is coordinated. A user should never be pressured to continue because stopping would supposedly reveal resistance, low intelligence or insufficient motivation.
Safe practice preserves autonomy
Use feedback only for a transparent, chosen purpose. Keep medical decisions with qualified clinicians. Maintain ordinary sleep, movement, learning and relationships. Stop when the cost, burden, adverse effects or lack of benefit outweigh the value. A technology that supports self-regulation should increase freedom—not make a person dependent on permission from a score.
Feedback can teach a skill; evidence must show what that skill changes
Neurofeedback and biofeedback are credible learning technologies, not one unified cure. EEG, fMRI, HRV, muscle and skin signals each reveal a narrow aspect of a living system. Some people learn to regulate those features. HRV biofeedback has encouraging evidence for selected emotional and physiological outcomes; specialized neurofeedback shows promising rehabilitation and mechanistic findings. For ADHD, PTSD, insomnia and broad cognitive enhancement, stronger controls often reduce or remove the apparent specific advantage.
The right response is neither ridicule nor technological faith. Define the target, verify the signal, test learning, demand transfer, measure function and preserve the person’s right to understand, refuse and leave. Used this way, feedback becomes a disciplined mirror—one that may help a mind and body practise change without pretending that the mirror created the person looking into it.
Sources and further reading
Core reviews, controlled trials, methodological standards and official regulatory or ethical documents supporting this guide. Evidence reviewed through September 3, 2026.
- Sitaram et al. Closed-loop brain training: the science of neurofeedback (2017).
- Ros et al. Consensus on the reporting and experimental design of clinical and cognitive-behavioural neurofeedback studies: CRED-nf checklist (2020).
- Galang, Velásquez, Elcin et al. Systematic review and meta-analysis of the relationships between real-time neurofeedback training parameters and acquisition of neural modulation (2025).
- Logothetis et al. Neurophysiological investigation of the basis of the fMRI signal (2001).
- Donoghue, Dominguez & Voytek. Electrophysiological frequency band ratio measures conflate periodic and aperiodic neural activity (2020).
- Westwood et al., European ADHD Guidelines Group. Neurofeedback for attention-deficit/hyperactivity disorder: A systematic review and meta-analysis (2025).
- Arnold et al., Neurofeedback Collaborative Group. Double-blind placebo-controlled randomized clinical trial of neurofeedback for ADHD with 13-month follow-up (2021).
- Wang et al. A double-blind randomized controlled trial of personalized upper-alpha neurofeedback in children with ADHD (2026).
- Lehrer & Gevirtz. Heart rate variability biofeedback: How and why does it work? (2014).
- Goessl, Curtiss & Hofmann. The effect of heart rate variability biofeedback training on stress and anxiety: A meta-analysis (2017).
- Lehrer et al. Heart rate variability biofeedback improves emotional and physical health and performance: A systematic review and meta-analysis (2020).
- Vann-Adibe et al. Efficacy and methodology of remote heart rate variability biofeedback interventions for mental health: A systematic review and meta-analysis (published 2025; volume 2026).
- Lalanza et al. Methods for heart rate variability biofeedback: A systematic review and guidelines (2023).
- Sumińska, Rynkiewicz & Szulczewski. Resonance frequency versus fixed 0.1 Hz breathing in HRV biofeedback: A four-week randomized comparison (2026).
- Berman et al. Systematic review and meta-analysis of neurofeedback training efficacy and neural mechanisms in the treatment of posttraumatic stress disorder (2025).
- Khaleghi et al. Effectiveness of fMRI-based neurofeedback therapy on depression: A systematic review (2025).
- Zhang et al. Self-navigating the “Island of Reil”: A systematic review of real-time fMRI neurofeedback training of insula activity (2025).
- Schabus et al. Better than sham? A double-blind placebo-controlled neurofeedback study in primary insomnia (2017).
- Butet et al. EEG-fMRI neurofeedback versus motor imagery after stroke: A randomized controlled trial (2025).
- Treves & Goldberg. Consumer-grade neurofeedback with mindfulness meditation: Meta-analysis (2025).
- US Electronic Code of Federal Regulations. 21 CFR § 882.5050—Biofeedback device (current electronic edition).
- US Food and Drug Administration. Product classification: Biofeedback device, product code HCC; and recognition of IEEE Std 2010-2023 for EEG neurofeedback systems (accessed 2026).
- European Parliament and Council. Regulation (EU) 2017/745 on medical devices (consolidated official text).
- European Parliament and Council. Regulation (EU) 2016/679: General Data Protection Regulation (official text).
- UNESCO. Recommendation on the Ethics of Neurotechnology (adopted 2025).
- OECD. Recommendation on Responsible Innovation in Neurotechnology (2019).
Educational note: This article explains general evidence, measurement and ethical principles. It does not diagnose a condition, interpret an individual EEG, prescribe breathing or neurofeedback treatment, or replace medical, psychological, rehabilitation or data-protection advice. Device software, intended use, personal health and local law materially change what is appropriate.
Altered states and cognitive enhancement series
- Flow States and Peak Performance
- Meditative States
- Sleep and Dreams
- Hypnosis and Suggestibility
- Psychedelic Research
- Neurofeedback and Biofeedback