The Brain in Front of the Mirror: How Neurofeedback Works and What We Know About the Brain’s Ability to Modulate Its Own Activity.
A dimly lit room. Small sensors are attached to a woman’s head; in front of her, a glowing circle expands and contracts. When a pattern of brain activity approaches the preset value, the circle expands and the music becomes clearer; when it strays from it, the image fades.
The woman is not observing her own thoughts: no machine can directly show her a memory, a fear, or a desire. She is receiving the translation, in images and sounds, of certain neurophysiological parameters. A process normally hidden from consciousness thus becomes perceptible and, through repeated experience, potentially modifiable.
This is the principle of neurofeedback: representing certain aspects of brain activity in real time to promote self-regulation.
Neurofeedback belongs to the broader family of biofeedback and uses signals from the central nervous system. The procedure creates a closed loop: brain activity is recorded, processed, and fed back in the form of images, sounds, or tactile stimuli.
The person thus receives information about their own neurophysiological variations and can modify their response either consciously or implicitly.
The most widely used technique is based on electroencephalography, or EEG. Electrodes placed on the scalp detect potential differences associated with the synchronized activity of large populations of neurons.
No current is introduced into the brain; the device records extremely weak electrical signals. The EEG does not observe individual neurons and does not read the contents of the mind. It records aggregate oscillations, analyzed based on frequency, amplitude, spatial distribution, and temporal evolution.
The bands conventionally defined as delta, theta, alpha, beta, and gamma indicate frequency ranges, not pure psychological states. Automatically equating alpha with relaxation or beta with attention oversimplifies complex phenomena: the meaning of each oscillation depends on the region involved, the task at hand, and the overall configuration of the trace.
Neurofeedback can also be performed using real-time functional magnetic resonance imaging (fMRI), which detects the BOLD signal associated with local changes in oxygenation and blood flow. Compared to EEG, it allows for the observation of deep regions and distributed networks with better spatial resolution, but it is costly and has lower temporal precision. Other methods include near-infrared functional spectroscopy, magnetoencephalography, and hemoencephalography, which provide indirect indicators of cortical hemodynamic and metabolic changes.
Regardless of the method, data acquisition is only the first step. Eye movements, muscle contractions, posture, and electrical interference can contaminate the data and must be identified as artifacts. After signal cleaning, the software calculates the predefined characteristic: the power of a frequency band, the ratio between multiple bands, the activation of a region, or the degree of functional connectivity between regions.
The value is compared to a threshold and transformed into a perceptible experience: the brightness of an image, the continuity of a melody, or the movement of a virtual object. For learning to occur, feedback must arrive with a latency compatible with the phenomenon being trained and must be genuinely linked to the selected parameter.
In conventional protocols, a parameter is identified that must be increased, decreased, or maintained within a range. The system rewards approaching the target and signals deviation from it.
Alongside these protocols, nonlinear dynamic neurofeedback has become widespread. This approach views the brain not as a simple, predictable mechanism, but as a complex and constantly changing system. A target state to be achieved is not necessarily defined: the algorithms analyze the EEG in real time and, when they detect a change deemed significant, introduce a very brief interruption in the sound stream. The micro-interruption is neither electrical stimulation nor a command, but sensory information synchronized with a variation in brain activity.
A person can listen to music or watch a movie without consciously employing a strategy. According to the theoretical model, the nervous system uses these signals to recognize its own transitions and spontaneously reorganize its functioning.
The most common metaphor is that of the mirror: the device does not tell the brain what to become, but rather provides it with a trace of its own change almost at the very moment it occurs. It is a striking image, but one that should be viewed with caution. The term “nonlinear” primarily describes the mathematical and conceptual model adopted; it does not prove that a technology fully represents the brain’s complexity, nor that every piece of information provided leads to self-regulation.
Another limitation concerns transparency. The algorithms of the leading commercial systems are not always fully described in peer-reviewed scientific literature. This makes it difficult to independently replicate the procedure and determine which components are responsible for any observed effects.
Research on nonlinear dynamic neurofeedback remains limited. Some pilot studies have reported subjective and neuropsychological improvements, but often in small samples and with study designs insufficient to demonstrate definitive clinical efficacy. There is a lack of independent, randomized, controlled trials using credible simulated procedures. The approach therefore constitutes an interesting hypothesis, not a technology already proven to be superior or universally effective.
Beyond the differences between methods, neurofeedback is not the same as simply observing an EEG. It is a dynamic process: the measurement alters the person’s experience, and this, in turn, influences what is detected. The traditional explanation draws on operant conditioning: when a neural state is followed by a rewarding event, the likelihood that the organism will reproduce similar conditions increases.
The process, however, is not reduced to a simple stimulus-response mechanism. Some participants use conscious strategies: they modify their breathing, direct their attention, evoke memories, or imagine movements. Others learn without being able to explain how. Explicit and implicit components can therefore coexist in the training.
Brain self-regulation is described as the acquisition of a skill involving attention, motivation, imaginative capacity, sensitivity to reward, and learning by trial and error. The person does not command neurons to produce a specific frequency; rather, they create mental and physical conditions in which certain states become more likely.
Individual differences are significant. The outcome depends on the protocol, the instrument, the form of feedback, and the ability to grasp—even implicitly—the relationship between experience and result. This variability renders a rigidly localizational conception insufficient: emotion, attention, memory, and cognitive control emerge from the interaction among distributed networks.
A randomized, double-blind study published in 2024, conducted on individuals with depression and persistent rumination, showed that the reduction in rumination was not explained by a simple linear change in the selected parameter. The improvement was associated with the interaction between regulatory effort, response to feedback, and the networks of frontal control, salience, and reward.
In some individuals, neurofeedback may therefore promote a different coordination among attentional, affective, motivational, and sensorimotor components.
The reference to neuroplasticity also requires clarification. Plasticity is the nervous system’s ability to change in response to experience, but it does not guarantee that every change is specific, stable, or clinically beneficial. To speak of consolidated learning, it must be verified that regulation improves over time, persists without external cues, and produces significant consequences in daily life.
The most rigorous protocols include transfer tests, in which the individual attempts to reproduce the learned state without real-time feedback. A change in the EEG during a session does not automatically equate to clinical improvement: electrophysiological changes may occur without functional effects, or subjective benefits may arise without the selected parameter changing in the expected direction. Neural learning, behavior, symptoms, and duration of effects must be evaluated separately.
Each session also involves nonspecific factors: the relationship with the practitioner, expectations, sustained concentration, motivation, repetition, and the perception of actively participating in treatment. These can produce real benefits, but they must be distinguished from effects attributable to brain modulation.
Without active controls, blinded procedures, and credible simulated feedback, it is difficult to determine which component caused the improvement. This problem emerged in a pre-registered, double-blind, controlled study published in 2026: alpha wave power increased in all groups, regardless of the authenticity of the feedback or conscious attempts at self-regulation.
Repetition, fatigue, and spontaneous fluctuations could explain at least part of the observed changes. The study involved a single session and does not rule out the efficacy of prolonged training, but it demonstrates how risky it is to interpret every EEG variation as evidence of specific learning.
Clinically, therefore, there is no single answer. The term neurofeedback encompasses interventions that vary greatly in terms of parameters, recording site, duration, session frequency, study population, and methodological quality.
In post-traumatic stress disorder, some reviews have found a reduction in symptoms, but the samples are often small and the protocols heterogeneous: the results are promising but not conclusive. For ADHD, a systematic review with meta-analysis published in JAMA Psychiatry in 2025, encompassing 38 randomized studies and 2,472 participants, found no significant improvements in core symptoms in blinded assessments. This finding does not rule out responsive subgroups or procedures worthy of further investigation, but it does not support the widespread use of neurofeedback as a standalone treatment.
Even for insomnia, a 2024 meta-analysis did not identify any substantial additional benefits compared to other psychological interventions or forms of biofeedback. In depression, chronic pain, neurological rehabilitation, and cognitive enhancement, interesting results have emerged, but the strength of the evidence varies considerably.
Overall, neurofeedback should be presented as a family of interventions still in the process of being defined, not as a universally effective therapy. The same caution is warranted regarding hypotheses about immune effects. The brain interacts with the autonomic nervous system, the hypothalamic-pituitary-adrenal axis, sleep, metabolism, and the immune response. It is plausible that improved stress management produces indirect physiological effects; however, demonstrating a change in an EEG rhythm or a reduction in a symptom is not equivalent to proving an anti-inflammatory effect.
It can be hypothesized that more effective regulation of attention and emotions reduces allostatic load, attenuates physiological hyperarousal, and promotes sleep, thereby indirectly influencing the autonomic, endocrine, and immune systems. Each step in this sequence, however, must be experimentally verified.
In clinical psychology, neurofeedback becomes more coherent when integrated into a comprehensive therapeutic plan. It can help individuals experience the variability of their internal states and strengthen their sense of personal efficacy. In some cases, it can facilitate psychotherapeutic work, especially when hyperarousal, dissociation, or attentional instability make it difficult to remain present in the experience and attribute meaning to it.
However, the technology does not automatically process trauma, does not alter relational patterns on its own, and does not reconstruct personal history. Modulating a neurophysiological parameter and understanding the meaning of one’s experience belong to different—albeit interconnected—levels. The protocol should therefore be based on a thorough clinical assessment, explicit functional goals, and continuous monitoring.
A quantitative EEG map can offer useful information, but it does not in itself constitute a psychiatric diagnosis, nor should it become an automatic catalog of anomalies to be corrected.
The future of this technique will depend less on the search for a universal model and more on the ability to identify which individuals can truly learn self-regulation, which indicators are clinically significant, and which feedback methods facilitate the transfer of these skills into daily life.
Adaptive devices, multivariate analysis, and the integration of EEG and neuroimaging may help describe distributed patterns rather than individual bands or isolated regions. The nonlinear approach, too, must be evaluated according to the criteria required of any clinical intervention: transparency of mechanisms, replicability, controlled comparisons, detection of adverse effects, and verification of the duration of benefits.
The complexity of the brain cannot be invoked to exempt a technique from scientific scrutiny; on the contrary, it must make the investigation more rigorous.
In the dimly lit room, the circle of light continues to expand and contract. The woman has not learned to control her own brain. Perhaps she has begun to recognize the conditions under which a particular mental configuration becomes more accessible; or her nervous system has received information about its own change and has attempted to integrate it.
Neither consciousness nor the secret of identity appears on the screen. Only a partial and measurable trace of the incessant dialogue between organism, experience, and environment emerges. Then the circle widens and the music resumes: not because the brain has finally obeyed, but because it may have recognized a variation in its own path and found a different way to continue along it.
