How FM-Theta Neurofeedback Learning Unfolds

A new mega-analysis by Enriquez-Geppert and colleagues brings welcome statistical heft to a question that has hovered over frontal-midline theta neurofeedback for years: not simply whether people can learn to upregulate FM-theta, but how that learning actually unfolds over time, how consistent it is across labs, and why some individuals seem to benefit more than others. Pooling raw participant-level data from five independent studies, the authors assembled what is, to date, the largest EEG-based FM-theta neurofeedback dataset of its kind.

That matters because FM-theta sits at an interesting intersection of mechanism and application. Frontal-midline theta activity is closely tied to executive control, conflict monitoring, and the recruitment of effortful top-down regulation. In principle, then, it offers a neurophysiologically grounded target for training people to better engage control-related networks. Biofeedback refers broadly to learning from real-time physiological signals, while neurofeedback is the EEG-based branch in which ongoing brain activity is measured and fed back so that participants can gradually learn to regulate it.

This paper is especially relevant for clinicians and neurofeedback professionals because it moves beyond simple pre/post outcome claims. Instead, it asks when learning appears, how stable it is, whether individualized frequency bands outperform the conventional 4–8 Hz theta range, and which participant characteristics may shape success. Those are not minor technical questions. They are exactly the questions that determine how we design protocols, how we interpret nonresponse, and how confidently we translate laboratory findings into real-world practice.

Methods

This study aggregated raw data from five FM-theta neurofeedback datasets collected in Groningen, Saarland, Toulouse, Valladolid, and Oslo, yielding a total sample of 168 participants. The sample included 94 females, 73 males, and one participant identified as other, with a mean age of 27.7 years. Most participants were healthy adults, although the Groningen dataset also included adults with self-reported executive-function complaints, and some of those participants reported diagnosed or suspected psychiatric conditions.

Across sites, participants were assigned to either a neurofeedback condition or an active control condition. Most studies were single-blind; the Oslo study was double-blind. Training protocols varied somewhat, but all shared a common FM-theta logic. Participants completed between six and eight sessions, with five or six neurofeedback blocks per session, and each block lasted five minutes. Feedback was delivered from either Fz alone or a frontal-midline montage including Fz, FC1, FC2, FCz, and Cz. Interfaces differed by study and included color-changing squares, a roller-coaster display, and gamified visual scenarios.

A particularly important methodological feature was the comparison between two theta definitions. The first was the standard FM-theta band of 4–8 Hz. The second was an individualized band centered on each participant’s own theta peak, defined as peak frequency ±1 Hz. Those peaks were estimated either from resting-state EEG or from task-related executive-function EEG, depending on the study. The authors then tested whether neurofeedback effects were specific to theta by comparing them against delta, alpha, and beta control bands.

To harmonize data across studies, amplitudes were transformed into within-session Z-scores relative to that session’s baseline. The authors examined learning in two complementary ways: a session-to-session index based on average performance across the first six common sessions, and a within-session index based on block-by-block changes across those sessions. They also calculated first-to-last session effects to capture the maximum training gain available within each study’s full protocol. Finally, they ran predictor analyses for neurofeedback success and exploratory responder analyses based on individual learning slopes.

Results

The central result is straightforward: participants receiving neurofeedback showed greater FM-theta upregulation than active controls. For standard FM-theta, the session-to-session analysis showed a significant main effect of group, with the neurofeedback group exhibiting overall stronger upregulation than the control group, F(1,158) = 17.81, p < .001, ηp² = 0.101. There was also a modest session effect, F(3.96,624.95) = 3.59, p = .007, ηp² = 0.022, but no significant session-by-group interaction, suggesting that the neurofeedback advantage appeared early rather than progressively widening over time.

The same general pattern held for individualized FM-theta, although with more heterogeneity. The group effect remained significant, F(1,158) = 12.51, p < .001, ηp² = 0.073, but the overall session effect did not reach significance, p = .074. In other words, individualized training still separated neurofeedback from control, but it did so less uniformly across studies.

Within-session analyses were arguably the most revealing. For standard FM-theta, the block-by-group interaction was significant, F(2.31,325.33) = 8.12, p < .001, ηp² = 0.054, alongside a strong main effect of block, F(2.31,325.33) = 38.39, p < .001, ηp² = 0.214. Individualized FM-theta showed a similar pattern, with a significant block-by-group interaction, F(2.75,435.07) = 8.46, p < .001, ηp² = 0.051. These findings suggest that the clearest signal of learning was expressed during active feedback blocks inside sessions, not as a neat session-by-session climb.

First-to-last session comparisons likewise supported a real neurofeedback effect. Group differences remained significant for both standard FM-theta, F(1,158) = 4.35, p = .039, ηp² = 0.027, and individualized FM-theta, F(1,158) = 6.54, p = .011, ηp² = 0.040, with reliable first-to-last increases across training. Importantly, robust group-level specificity was largely confined to theta. Delta showed some training-related change but not neurofeedback-specific change; alpha and beta effects were inconsistent and study-dependent, although beta showed transient within-session modulation in some analyses.

Predictor analyses found that female sex and lower educational attainment were associated with greater neurofeedback success in the standard theta band. In the adjusted model, sex remained significant, β = 0.183, p = .013, and high education predicted lower success, β = -0.359, p = .003. These effects should be interpreted cautiously, however, because some sample characteristics were entangled with study membership. Exploratory responder analyses showed that 72.7% of the neurofeedback group met the responder criterion for standard FM-theta, versus 27.3% classified as non-responders. Yet 65% of the active control group also met the same responder criterion, which immediately complicates any simplistic interpretation of “response” as proof of feedback-specific learning. Non-responders in the neurofeedback group more often reported or suspected psychiatric disorders.

Discussion

This paper adds something unusually valuable to the neurofeedback literature: a more realistic picture of what EEG learning looks like when one stops expecting it to behave like a staircase. The dominant pattern here was not a steady session-by-session climb in FM-theta amplitude. Instead, the neurofeedback group gained an early advantage over active controls and then largely maintained that advantage. The strongest learning signal showed up within sessions, during the actual feedback blocks, which suggests that FM-theta self-regulation may be acquired relatively quickly and then stabilized through practice.

That distinction matters. In many clinical and performance settings, progress is implicitly judged by whether the signal keeps rising from one session to the next. This mega-analysis suggests that such an expectation may be mismatched to the phenomenon. For FM-theta neurofeedback, the better question may be whether the person can reliably enter and sustain the target state during training, and whether that ability becomes more reproducible across days, rather than whether the raw theta value climbs monotonically each session. From a protocol-design perspective, this supports the use of within-session learning markers and early-session checkpoints rather than waiting for a dramatic late-session ramp that may never come.

The study also helps disentangle two related but distinct issues: electrophysiological self-regulation and downstream behavioral change. The present analyses were about the former. They do not establish that early FM-theta upregulation automatically translates into clinical or cognitive benefit on the same timetable. In fact, the authors are appropriately careful here. It is quite plausible that learning to modulate the signal happens relatively early, whereas transfer to executive functioning, daily life, or symptom reduction requires additional consolidation, task practice, expectancy alignment, or ecological relevance. For clinicians, that means shorter protocols may be sufficient to establish self-regulation capacity, but not necessarily sufficient to produce meaningful real-world change.

The contrast between standard and individualized FM-theta is especially interesting. On paper, individualized peaks are appealing because frontal theta frequency varies across people and may be stable within individuals. In practice, however, the individualized band produced more heterogeneous, study-dependent effects than the standard 4–8 Hz band. That does not invalidate individualized training. It may simply indicate that individualized targeting is highly sensitive to how peaks are derived, whether they come from rest or task performance, and whether the underlying executive tasks recruit partially different theta generators. Clinically, that suggests a useful caution: personalization is not automatically superior merely because it is personalized. Sometimes the more standard target captures the shared network phenomenon more robustly, while individualized approaches may require tighter measurement and better operational consistency.

The responder findings are also worth treating with nuance. A sizeable majority of the neurofeedback group met the responder definition, but so did a surprisingly large fraction of active controls. That tells us two things. First, mental strategy, engagement, and expectancy can move FM-theta even without veridical feedback. Second, responder classification based on slopes alone is not the same thing as demonstrating feedback-specific efficacy. At the same time, the descriptive overrepresentation of psychiatric complaints in non-responders is clinically provocative. It is not causal evidence, especially because diagnostic status was confounded with one study site, but it does raise the possibility that attentional dysregulation, altered reward processing, fatigue, or network-level differences in medial frontal function could meaningfully affect trainability.

From a translational standpoint, this paper nudges the field toward a more adaptive model of practice. Rather than assuming everyone should follow the same dose and the same target band for the same number of sessions, it supports monitoring early learning signatures, adjusting strategy coaching, and being cautious about labeling someone a non-responder too quickly or too rigidly. It also strengthens the argument that theta training is not merely producing nonspecific broadband spectral drift, since the cleanest and most reproducible group effects were concentrated in the theta range. That point deserves emphasis in the broader neurofeedback debate. Placebo-style influences, expectancy, motivation, and task engagement can certainly affect training behavior and even shape physiology to some extent; this paper does not make those factors disappear. But the combination of active-control comparison and band-specific effects makes a purely placebo account much harder to sustain. If the observed changes were mainly generic enthusiasm, demand characteristics, or simple exposure to the training environment, one would expect broader, less selective shifts across bands and weaker separation from controls. Instead, the most reliable effects clustered around the trained theta target. That is not proof that every clinical benefit from neurofeedback is feedback-specific, but it is solid evidence that the electrophysiological learning signal itself cannot be dismissed as just placebo.

At the same time, the study’s limits matter. The included protocols were heterogeneous, blinding was not uniform, several clinically interesting variables were inseparable from site effects, and behavioral outcomes were not harmonized. So this is not the last word on clinical efficacy. But it is an important methodological advance. It shifts the conversation from “does FM-theta neurofeedback work at all?” toward a more useful set of questions: when does it work, how do we know learning is occurring, and what design choices make that learning more reproducible and more likely to matter outside the training chair?

Brendan's perspective

What I appreciate most about this paper is that it speaks to a frustration many clinicians quietly carry: we know neurofeedback learning is often real in the room, but the research literature has sometimes struggled to describe how that learning shows up in a way that looks clinically familiar. This mega-analysis helps by showing that FM-theta learning can be specific, early, and uneven across people without becoming meaningless. To me, that is a very practical message.

1. Standard first, individualize with a reason

If I were building an EEG neurofeedback protocol inspired by this paper, I would not walk away thinking individualized FM-theta is a bad idea. I would walk away thinking it should be earned. The standard 4–8 Hz target was more robust across studies, which tells me that in everyday practice the conventional range is still a very reasonable starting point when the clinical goal is to support executive control, sustained effort, or top-down regulation.

That matters because individualized frequency work sounds elegant, and sometimes it is elegant, but elegance is not the same as reliability. In real clinics, individualized theta peaks can shift depending on whether they are measured at rest, during a task, with eyes open, with eyes closed, or on a day when the client is tired, anxious, or under-caffeinated. If we personalize too quickly, without stable measurement and a clear rationale, we may end up tailoring the protocol to noise rather than to a meaningful physiological trait.

So my own bias would be to begin with a fairly clean frontal-midline setup, typically emphasizing Fz or a midline montage around Fz and FCz, and train standard FM-theta first when the case formulation supports it. Then I would individualize only if the session data give me a reason: unstable reward, very weak task engagement, a poorly expressed theta profile, or a client whose cognitive state suggests that the generic band is missing the mark. In other words, personalization should solve a problem, not decorate a protocol.

This is especially relevant when working with clients who present with attentional inefficiency, cognitive fatigue, or executive-control complaints. It can be tempting to over-customize from day one. But sometimes the better clinical move is to start with the most reproducible target, watch what the brain actually does, and let the first few sessions teach you how much customization is truly needed. The paper gives some support to that approach. Standardization, used thoughtfully, is not the enemy of individualized care. Often it is the platform that makes individualized care more accurate.

2. Early learning changes how I judge a protocol

The second idea I would carry directly into practice is that FM-theta learning may show up earlier than many of us expect, and that we may need to judge progress differently because of it. I think this is one of the most clinically useful parts of the paper. Many practitioners, and certainly many clients, assume learning should look like a nice clean staircase: session one is modest, session two is better, session three is better again, and so on. Brains, inconveniently, do not always read that script.

What this study suggests is that the important sign may be whether the person can enter the target state during active training blocks and do so with increasing consistency, not whether every session average rises in a linear way. That is a subtle but important shift. It means that by session two, three, or four I am already asking: can this client find the state? Can they re-enter it after losing it? Does the task feel more graspable? Is the reward signal becoming less chaotic? Those may be more useful questions than simply asking whether the mean theta value is still climbing.

From a protocol standpoint, that makes me more comfortable making earlier adjustments. If there is no sign of state access at all, I would not just keep repeating the same training out of hope or ritual. I would look at arousal level, fatigue, strategy coaching, feedback sensitivity, reward thresholding, and block duration. Some clients do better with shorter blocks and more resets. Some need less visual clutter and less performance pressure. Some need a brief period of breathing regulation, paced exhalation, or heart-rate-variability training before frontal theta work becomes trainable. None of that undermines EEG neurofeedback. It is part of creating the conditions in which EEG learning can actually occur.

Just as importantly, early success does not mean the job is done. I would take this paper as evidence that signal regulation can emerge early, but transfer may still take time. So if the client shows clear in-session access to FM-theta, I would usually continue long enough to stabilize that skill and connect it to daily-life demands. Laboratory learning and life learning are cousins, not twins. The protocol should respect both.

3. Variability is not failure, especially in complex cases

The responder data in this paper are messy, and I mean that as a compliment. Real clinical work is messy. Not everyone learns at the same pace, not everyone responds for the same reason, and not every apparent responder is demonstrating a fully feedback-specific effect. That complexity is exactly why this paper feels useful rather than naïve.

For ADHD, executive dysfunction, and more complex presentations, my takeaway is not that FM-theta training should be used indiscriminately. It is that variability should be expected, tracked, and interpreted with humility. A client with ADHD may have difficulty sustaining the exact kind of internally organized effort the protocol is asking for, at least initially. A client with trauma, mood instability, sleep disruption, or broader psychiatric complexity may show inconsistent engagement because the obstacle is not just “learning the signal.” It may be state regulation, reward sensitivity, mental fatigue, interoceptive tolerance, or simple cognitive overload.

That means I would be careful about labeling someone a non-responder too early. I would rather ask what is blocking learning. Is the target appropriate? Is the person overactivated? Underactivated? Trying too hard? Dissociating a little? Bored? Confused by the feedback metaphor? Clinically, those questions often lead to much better decisions than the binary question of whether the client “can do neurofeedback.”

This is also where broader treatment integration matters. For some clients, FM-theta may fit best after groundwork in sleep stabilization, autonomic regulation, medication consistency, psychotherapy, or basic self-monitoring. For others, it may be a strong early intervention precisely because it gives them a felt sense of focused control. The paper does not settle those sequencing questions, but it reminds us that interindividual variability is not a nuisance variable to be ignored. It is part of the treatment landscape.

So my overall reaction is optimistic. This study gives us better evidence that FM-theta learning is not just placebo in a lab coat, and it also gives us permission to be more intelligent about how we monitor it. Use a stable target before chasing hyper-personalization. Look for early state access, not just slow upward drift. And when variability appears, read it clinically rather than morally. The brain is not being stubborn. It is giving us information.

Conclusion

This mega-analysis makes a strong case that FM-theta neurofeedback is a real and reproducible electrophysiological training phenomenon, at least at the level of signal regulation. Across five independent datasets, participants receiving neurofeedback showed stronger FM-theta upregulation than active controls, and the clearest evidence of learning appeared within sessions rather than as a simple linear rise across sessions. That is a useful reframing for both research and practice.

Just as importantly, the paper shows that not all apparent learning patterns mean the same thing. Standard FM-theta was more robust at the group level than individualized theta, non-theta bands did not show the same reproducible pattern, and the neurofeedback group outperformed active controls on the trained signal. Taken together, that pattern is difficult to reconcile with a purely placebo explanation. Expectancy and nonspecific factors still matter, and the responder analyses make that very clear, but this study offers some of the better evidence to date that FM-theta neurofeedback involves target-specific electrophysiological learning rather than just a generic training artifact. In other words, the field is getting a more precise map of what FM-theta training does, what it does not yet prove, and where personalization may genuinely help.

For neurofeedback professionals, that is encouraging news. We do not need a perfect or final model to learn something meaningful from this paper. What we have here is a clearer signal that FM-theta can be trained, that learning may emerge earlier than many assume, and that future progress will likely come from smarter measurement, better adaptation, and more thoughtful matching between protocol and person.

References

Enriquez-Geppert, S., Smit, D., Eschmann, K. C. J., Marcos-Martínez, D., Hornero Sanchez, R., Hsieh, S., Dehais, F., & Huster, R. J. (2026). A mega-analysis of EEG-based frontal-midline theta neurofeedback reveals learning dynamics, individual variability, and response profiles. NeuroImage, 329, 121820. https://doi.org/10.1016/j.neuroimage.2026.121820