Studies report behavioural and cognitive improvements in children, but the number of randomised trials remains too small to call efficacy established; in adults there is almost no data.
Updated :
What the research shows
EEG neurofeedback — standard amplitude training
Mu/SMR at C4, low beta, theta-down, theta/beta ratio, alpha; qEEG-guided protocol selection included
Ages
Child (under 12)
Techniques
Sensorimotor rhythm (SMR), Beta training, Theta/beta ratio, Alpha training
AAPB3Probably efficaciousNeuroLogic3Probably efficacioussame vs AAPB
Level 3 maintained. The AAPB base is paediatric: Jarusiewicz 2002 (N = 40, half randomised to waitlist, 12 completers averaging 36 sessions of 10-13 Hz up-training at C4), Kouijzer 2009 (n = 14, 40 sessions of low-beta up / theta down at C4, executive-function and parent-rated gains persisting at follow-up), Kouijzer 2010 (n = 20, qEEG-guided theta suppression, social behaviour improved) and Pineda 2008 (randomised, participants blinded, 30 sessions of mu training at C4 with fake mu feedback in the placebo arm: mu changed only in the active arm, gains in sustained attention and sensory and cognitive awareness). Three randomised trials add to this since 2022: Wang 2024 (n = 60, ages 3-6, 60 sessions of wearable mu-rhythm neurofeedback vs sham at two centres, both arms on similar behavioural programmes; greater gains in expressive language, P = 0.013, and cognitive awareness including joint attention, P = 0.003), Kang 2026 (n = 40 preschoolers, 4-week alpha protocol vs sham on an 8-channel portable system; ABC total and Social Relating reduced) and Li 2026 (n = 40, alpha training vs no feedback; caregiver-rated SRS/ABC social cognition and relating improved). Held at 3 rather than raised to 4: the new trials come from overlapping groups on low-density portable systems without stated assessor blinding, and the older randomised evidence is very small.
Mu/SMR at C4, low beta, theta-down, theta/beta ratio, alpha; qEEG-guided protocol selection included
Ages
Adolescent (12-17)
Techniques
Sensorimotor rhythm (SMR), Beta training, Theta/beta ratio, Alpha training
AAPB3Probably efficaciousNeuroLogic3Probably efficacioussame vs AAPB
Level 3 maintained. The one full randomised controlled trial in the AAPB base is adolescent: Kouijzer 2013, n = 38 aged 12-18, randomised to neurofeedback (n = 13, 40 sessions at Cz or FCz, training target blinded to trainees), skin-conductance biofeedback (40 sessions) or waitlist; participants who achieved the training goals ('regulators', a subgroup analysis) improved in cognitive flexibility, maintained at 6-month follow-up and not found in the biofeedback or waitlist groups. Kouijzer 2010 (ages 8-12) and the AAPB's mixed-age cohorts border this band. Since 2022 the best-powered controlled neurofeedback trial in ASD is adolescent — the Vienna slow-cortical-potential trial (n = 41, 24 sessions vs treatment as usual; Klöbl 2023, Fietz 2025, Auer 2025) — and it is null on clinical and affective outcomes while showing protocol-related EEG change; SCP is not amplitude training and has its own row, which neither supports nor lowers this one. Esmaeilzadeh Kanafgourabi 2025 (ILF, n = 24) likewise sits in its own row. Level 3 rests on one small randomised trial with a subgroup result and on continuity with the child evidence; no adolescent trial is adequately powered.
Mu/SMR at C4, low beta, theta-down, theta/beta ratio, alpha; qEEG-guided protocol selection included
Ages
Adult (18+)
Techniques
Sensorimotor rhythm (SMR), Beta training, Theta/beta ratio, Alpha training
AAPB3Probably efficaciousNeuroLogic2Possibly efficaciouslower vs AAPB
Level 2 (the AAPB level rests on paediatric studies). AAPB states that 'the majority of studies were done on pediatric populations' and its randomised and quasi-experimental base is entirely under 18. No controlled trial of clinician-delivered amplitude neurofeedback in autistic adults exists in the AAPB base or in the 2022-2026 window. The only adult-inclusive controlled record is Brewe 2025 (n = 27 autistic males, mean age 21.12, ten sessions of mixed-reality EEG brain-computer-interface feedback vs waitlist: a group difference in facial emotion recognition at endpoint, but no reliable individual change), which is neither amplitude training nor a symptom outcome; Coulter 2022 includes adults but is HRV biofeedback (see the biofeedback box). Thompson 2010's clinical series (150 people with Asperger's syndrome, mostly children and adolescents) is the remaining base. The band is shown so that the adult picture is not read off the paediatric level: the training logic (mu/SMR at C4, theta-down) is the same, but an age band with no literature of its own cannot be rated above 2.
AAPB3Probably efficaciousNeuroLogic3Probably efficacioussame vs AAPB
Level 3 (the AAPB level is a pooled rating across all neurofeedback protocols; coherence training contributes one group's dataset to it). Coben & Padolsky 2007: 37 children trained with connectivity-guided neurofeedback and 12 children with similar characteristics as waitlist controls; parents reported significant improvements in overall ASD symptoms and executive functioning, with neuropsychological gains in attention, visual-perceptual functioning, executive function and language. Coben 2014 reports connectivity change accompanied by behavioural improvement in what appears to be the same 37/12 sample; Coben & Myers 2010 compared connectivity-guided with symptom-based training. The design meets Level 3 on its face (waitlist-controlled, replicated within the cohort), but it is one research group, non-randomised assignment, no independent replication in almost twenty years and no 2022-2026 record. Kang 2026 reports alpha high-order connectivity change correlating with symptom change (r = 0.678, p = 0.015), but as a mechanism of amplitude training, not as connectivity training. The connectivity literature is strongest in network-characterised conditions such as ASD, yet remains weakened by under-controlled designs and by the measurement basis of surface connectivity.
Slow cortical potentials (SCP)
Ages
Adolescent (12-17)
Techniques
Slow cortical potentials (SCP)
AAPB0Not ratedNeuroLogic2Possibly efficacious
Level 2. One adequately described randomised trial, in adolescents, and it did not separate from its control: the Vienna trial (DRKS00012339), 24 sessions of prefrontal slow-cortical-potential training vs treatment as usual, n = 41 (21 neurofeedback / 20 treatment as usual). Klöbl 2023: no significant group × time interaction for affective or resting-state measures; SCP neurofeedback 'did not lead to superior improvements in neuronal or affective functioning compared to treatment as usual', although activation corresponding to SCP differentiation correlated with affective improvement. Fietz 2025: a group × time interaction in P300 latency (shorter after neurofeedback, longer in controls) with non-significant post hoc tests and no clear empathy advantage. Auer 2025: alpha rose in the neurofeedback group and fell in controls, the alpha increase being linked to enhanced positive affect. Prillinger 2022: impulsivity moderated the CNV response. Protocol-specific EEG change with clinical fidelity (clinician-delivered, 24 sessions), but no demonstrated clinical superiority and no second trial. Under the framework a treatment-as-usual null is a real negative on symptoms, read with its limits (no sham, n = 41, male-only in the fMRI report).
Infra-low frequency (ILF) / infra-slow (ISF)
Ages
Child (under 12) · Adolescent (12-17)
Techniques
Infra-low frequency (ILF)
AAPB0Not ratedNeuroLogic2Possibly efficacious
Level 2. Esmaeilzadeh Kanafgourabi 2025: single-blind randomised trial, n = 24 high-functioning adolescents (12 per arm), 15 one-hour sessions of ILF neurofeedback vs an inactive sham; inhibitory control and central alpha, theta and gamma absolute power improved at post-test, with no significant effect at follow-up on either behavioural or EEG measures. Saleem 2024: single-arm study, n = 35 children aged 7-17, 30 sessions over 10 weeks; NIH Toolbox inhibitory control, cognitive flexibility, processing speed and working memory improved (p = 0.00), with a trend of improvement at 2-month follow-up and no control group; the companion EEG paper (Saleem 2024, Pak J Med Sci) reports reduced delta, theta and alpha. One small controlled study whose gains did not persist and one uncontrolled cohort that cannot exclude practice effects; 15 sessions is below the fidelity bar and there is no independent replication. Cognitive rather than core-symptom outcomes throughout.
Real-time fMRI neurofeedback
Ages
Adult (18+)
Techniques
fMRI neurofeedback
AAPB0Not ratedNeuroLogic2Possibly efficacious
Level 2. Pereira 2024: the first real-time fMRI neurofeedback study in ASD, a working-memory paradigm targeting the left dorsolateral prefrontal cortex in 13 autistic individuals without intellectual disability and 17 neurotypical controls. Both groups modulated the target region (84 % and 98 % of runs); target activity remained lower in the ASD group, particularly without feedback, with compensatory connectivity changes between the DLPFC and motor, visual and multiple-demand regions in transfer runs. Mechanistic feasibility and target engagement only — no clinical endpoint, no clinical control group, no follow-up. The row is shown because a study exists; nothing clinical can be inferred from it yet.
Biofeedback
HRV biofeedback — anxiety and stress in ASD
Ages
Adolescent (12-17) · Adult (18+)
Techniques
HRV — resonance-frequency breathing
AAPB0Not ratedNeuroLogic2Possibly efficacious
Level 2. The AAPB did not rate biofeedback for ASD ('the current review is based only on neurofeedback studies outcomes'): this row is new rather than a disagreement. Thoen 2024: single-blind randomised sham-controlled pilot, n = 44 autistic adolescents (24 supervised HRV biofeedback / 20 sham), followed by an optional home phase — a late increase in cardiac vagal modulation, heart rate up and cortisol down immediately after supervised training but not at follow-up, lower self-reported stress after home practice, no significant change in psychosocial functioning, and a significant fall in compliance at home. Coulter 2022: home-based pilot, n = 20 aged 13-24, 12 weeks of self-managed devices; pre-post anxiety reduced in the under-18s (t6 = 2.55, P = .04, d = 0.99) and in the adults (t7 = 3.95, P = .006, d = 0.54), uncontrolled. Coulter 2024 (PRISMA mini review): positive short-term effects on anxiety, no long-term follow-up, no pooled estimate. One adequately controlled pilot with physiological but not psychosocial change, and one uncontrolled pilot: Level 2, pooled across adolescents and young adults.
By age
Child (under 12)
In children, neurofeedback has been tested in about ten small studies and, since 2022, in three randomised trials. Trained children improve in attention, language and social behaviour according to parents and teachers, and the recent trials against a sham training point the same way. Studies remain small, often without an independent assessor, and it is not known whether the gains last.
AAPB Level 3 confirmed by NeuroLogic. Pineda 2008 (randomised, EMG placebo + fake mu), Jarusiewicz 2002, Kouijzer 2009/2010 (waitlist, C4, 40 sessions); Wang 2024 (n = 60, sham, expressive language and joint attention), Kang 2026 and Li 2026 (alpha, n = 40 each). Held at 3: overlapping groups, portable systems, assessor blinding not stated.
Adolescent (12-17)
In adolescents, one Dutch randomised trial shows a cognitive-flexibility gain maintained at six months in those who learned to regulate their EEG. The largest recent trial, in Vienna, tested a different form of neurofeedback (slow cortical potentials) against usual care: the EEG changed as expected, but symptoms did not improve more than in controls.
NeuroLogic Level 3 (AAPB 3 pooled child/adolescent). Kouijzer 2013 (n = 38, ages 12-18, NF vs SCL biofeedback vs waitlist; subgroup result in regulators). Vienna SCP trial (n = 41, 24 sessions vs treatment as usual): null on clinical outcomes, alpha increased (Klöbl 2023, Fietz 2025, Auer 2025) — rated in the SCP row.
Adult (18+)
In autistic adults, neurofeedback has hardly been studied: no controlled trial of standard training exists. One small American trial of a brain-computer-interface-assisted programme and one home-based heart-rate biofeedback pilot are the only data. The level assigned in children cannot be carried over.
NeuroLogic Level 2 (the AAPB Level 3 rests on paediatric studies). Brewe 2025 (n = 27, mixed-reality BCI vs waitlist, facial emotion recognition, no reliable individual change); Coulter 2022 (home HRV, ages 13-24, d = 0.54 in adults, uncontrolled); Pereira 2024 (fMRI, mechanistic).
In short
Clinical reading
NeuroLogic Level 3 for amplitude neurofeedback in children and adolescents (AAPB 3 maintained: small RCTs by Kouijzer 2013 and Pineda 2008, waitlist-controlled quasi-experimental studies, three new Chinese RCTs against sham or no feedback), 2 in adults for want of any controlled trial. Coherence (Coben) 3 on one group's cohort; SCP 2 (Vienna trial vs treatment as usual, null on symptoms); ILF 2; fMRI 2; HRV biofeedback 2 (not rated by the AAPB). Most work is quasi-experimental or case series; protocols closely resemble ADHD protocols, which raises questions about target specificity.
Protocols
Mu/SMR up-regulation with theta and high-beta inhibition at C4 (Pineda, Kouijzer), theta-down or qEEG-guided training at Cz/FCz, alpha training in recent trials; connectivity-guided training as a separate approach; 30-40 sessions.
Limits
Few RCTs, heterogeneous protocols and outcome measures, no long follow-up; outcomes mostly cognitive and parent-rated rather than core symptoms; the one well-conducted trial against treatment as usual (SCP, Vienna) is null on symptoms; no controlled trial in autistic adults; the heterogeneity of ASD itself limits generalisation.
Study base
AAPB base: two small RCTs (Kouijzer 2013, Pineda 2008) and about ten quasi-experimental studies; 2022-2026 base: 13 publications indexed in the archive (11 neurofeedback, 2 HRV biofeedback), including three new sham- or no-feedback-controlled RCTs.
Brendan's perspective
Level 3 in children and adolescents, same as the AAPB, and 2 in adults because there is no adult literature to read: Brewe 2025 is a mixed-reality BCI feasibility study, not amplitude training, and Thompson 2010 is a clinical series. I hold the coherence row at 3 with the caveat I would apply to any method: one group, one cohort (Coben & Padolsky 2007), non-randomised assignment, parent-rated outcomes, and nobody independent has reproduced it in almost twenty years. That absence is itself a finding. ASD is a substantial part of my practice. Most of that work is mu/SMR at C4 with theta down, protocol derived from the qEEG rather than a menu, 30 to 40 sessions — protocols that look a great deal like the ADHD ones, which should make everyone ask what is actually being targeted. The other approach I find genuinely interesting is inhibiting anterior bilateral delta and theta: on the left that territory is tied to communication, on the right to social function, and both sides to attention and executive control generally. That is a rationale, not a result, and I hold it loosely. The honest limit: the best-controlled trial here, the Vienna SCP trial against treatment as usual, is null on symptoms (Klöbl 2023), and most reported gains are cognitive and parent-rated. Some of that is the design problem rather than the method. Running an adequately controlled trial in an autistic sample is genuinely hard — heterogeneous presentations, outcomes that lean on parent report, and a sham condition that asks a great deal of the participants it is meant to blind. As I argued in Parsons 2026, none of that is inherently a blocker to demonstrating efficacy; it does mean this population needs research designs built for it rather than borrowed from elsewhere.
Kouijzer et al. (2013) Is EEG-biofeedback an effective treatment in autism spectrum disorders? A randomized controlled trial doi:10.1007/s10484-012-9204-3
Coben & Myers (2010) The relative efficacy of connectivity guided and symptom-based EEG biofeedback for autistic disorders doi:10.1007/s10484-009-9102-5
Pineda et al. (2008) Positive behavioral and electrophysiological changes following neurofeedback training in children with autism doi:10.1016/j.rasd.2007.12.003
Coben et al. (2014) Connectivity-guided EEG biofeedback for autism spectrum disorder: Evidence of neurophysiological changes doi:10.15540/nr.1.2.109
Wang et al. (2024) Wearable EEG Neurofeedback Based-on Machine Learning Algorithms for Children with Autism: A Randomized, Placebo-controlled Study doi:10.1007/s11596-024-2938-3
Kang et al. (2026) High-order alpha-band network reorganization underpins clinical symptom alleviation following EEG neurofeedback in preschool children with autism spectrum disorder: a multidimensional study doi:10.3389/fnhum.2026.1897176
Klöbl et al. (2023) Individual brain regulation as learned via neurofeedback is related to affective changes in adolescents with autism spectrum disorder doi:10.1186/s13034-022-00549-9
Thoen et al. (2024) The Physiological and Clinical-Behavioral Effects of Heart Rate Variability Biofeedback in Adolescents with Autism: A Pilot Randomized Controlled Trial doi:10.1007/s10484-024-09638-1
