The Six Types of Neurofeedback (And Why They're Not All the Same)

Not All Brain Training Is Created Equal

Imagine you've decided to learn a musical instrument. You walk into a music school, and the receptionist says, "Great! We teach music here." Then she hands you a trombone, a sitar, an electronic drum pad, a theremin, a pipe organ, and a didgeridoo.

"These are all music," she says. "Pick one."

You'd probably want to know how they differ before choosing. What kind of music each one makes. How hard each one is to learn. Whether any of them require a 200-year-old cathedral to play properly.

This is roughly the situation anyone encounters when they start researching neurofeedback. The word gets thrown around as if it describes a single thing. It doesn't. There are at least six fundamentally different types of neurofeedback, and they vary so dramatically in mechanism, equipment requirements, evidence base, and clinical application that lumping them together under one label borders on misleading.

Some train specific brainwave frequencies at the scalp surface. Some use mathematical models to target structures buried deep inside the brain. One doesn't even use EEG at all. It measures blood flow with infrared light.

Here's a map of the whole landscape so you can actually make sense of it.

The Common Thread: A Feedback Loop Aimed at the Brain

Before we split neurofeedback into its various species, it's worth understanding the genus. Every type of neurofeedback shares one fundamental principle: measure something about brain activity, display that measurement in real time, and let the brain learn to change it through operant conditioning.

That's it. That's the shared DNA.

The differences are all about what you measure, where you measure it, and how you feed the information back. And those differences turn out to matter enormously.

Think of it this way. A cardiologist and a gastroenterologist both practice medicine. They both diagnose and treat. But what they measure, where they look, and what they do about what they find are so different that calling them both "doctors" only tells you so much.

Same story here. Let's meet the family.

Type 1: Frequency and Amplitude Training (The Original)

This is the one that started it all, and it's still the most widely practiced type. When most people say "neurofeedback," this is what they mean.

The idea is beautifully simple. Your brain produces electrical oscillations at different frequencies, from slow delta brainwaves during deep sleep to fast gamma bursts during moments of insight. These frequency bands correspond to different mental states. Too much slow-wave theta activity over the frontal cortex? That's associated with inattention and mind-wandering. Too much fast high-beta activity? That often correlates with anxiety and rumination.

Frequency training picks a specific band at a specific location and either rewards your brain for producing more of it (uptraining) or rewards it for producing less (downtraining). A classic ADHD brain patterns protocol, for instance, might uptrain beta (13-21 Hz) and downtrain theta (4-8 Hz) at electrode site Cz (top of the head). The goal: shift the theta-to-beta ratio toward what's typical for focused attention.

The feedback itself can be anything perceivable. A video that plays smoothly when you hit the target. A tone that sounds when your brain drifts. A game character that moves when your brainwaves cooperate. The form doesn't matter nearly as much as the contingency: immediate reward when the target pattern appears, immediate withdrawal when it doesn't.

The evidence

Frequency training has the deepest evidence base of any neurofeedback type. The research on ADHD alone spans decades. A 2019 meta-analysis in European Child and Adolescent Psychiatry (Cortese et al.) found significant improvements in inattention that held at follow-up. The American Academy of Pediatrics rates it Level 1 ("Best Support") for ADHD. Studies also support its use for anxiety, insomnia (particularly SMR training at 12-15 Hz), and focus optimization in healthy populations.

The equipment

At minimum, you need one or two EEG channels, a sampling rate of at least 256 Hz, and software that can extract band power in real time. But single-channel setups are limiting because they can't distinguish between brain regions. An 8-channel device like the Neurosity Crown covers frontal, central, and parietal sites, which means you can run multi-site protocols that target specific cortical networks rather than just "the brain" in general.

Type 2: Z-Score Neurofeedback (Training Toward Normal)

Here's where things get more sophisticated.

Frequency training has a limitation: somebody has to decide what frequency to train and in which direction. A clinician looks at a QEEG brain map, identifies deviations from a normative database, and selects a protocol. It works, but it's manual. It trains one or two metrics at a time. And it requires an expert to interpret the data and design the training plan.

Z-score neurofeedback automates a big chunk of that decision-making. Instead of training a single frequency band, it simultaneously compares multiple metrics of your brain activity, power, asymmetry, coherence, phase, and more, against a normative reference database. Your brain activity at each site is converted to Z-scores (standard deviations from the database mean). The feedback then rewards your brain for moving all of those Z-scores closer to zero. Closer to the statistical center of "normal."

Think of it like this. Frequency training is a personal trainer who says, "Do more bicep curls." Z-score training is a personal trainer with a full-body scan who says, "Here are the 47 ways your body deviates from optimal. Let's move everything toward the center, simultaneously."

The attraction

The appeal of Z-score neurofeedback is efficiency. Because it trains multiple metrics at once, proponents argue it can produce results in fewer sessions than single-band frequency training. Some clinicians report noticeable changes in 10 to 15 sessions rather than the 30 to 40 typical for ADHD frequency protocols. The approach also reduces the reliance on clinician expertise in protocol selection because the normative database guides the training automatically.

The controversy

Critics raise a fair point: who says "normal" is optimal? A normative database represents the statistical average of a population. But some deviations from average might be functional. An artist with unusual right-hemisphere coherence patterns might lose something if those patterns are pushed toward the population mean. The Z-score approach assumes that closer to average equals healthier, and that's not always true.

The evidence base is growing but still smaller than frequency training's. Several controlled studies show promising results, particularly for attention disorders and anxiety, but more large-scale randomized trials are needed.

The equipment

Z-score training works best with at least 4 channels (to compute meaningful inter-site metrics like coherence and asymmetry) and access to a normative database. The Neurosity Crown's 8 channels covering all major cortical lobes make it well suited for this approach, particularly when paired with open-source tools that include normative reference data.

Type 3: LORETA Neurofeedback (Going Deep)

This is where neurofeedback gets genuinely mind-bending.

Standard surface EEG neurofeedback, whether frequency-based or Z-score, trains the electrical activity you can detect at the scalp. But the scalp is the outside of your head. A lot of interesting brain activity happens in structures that sit inches below the surface: the anterior cingulate cortex (involved in error monitoring and emotional regulation), the insula (central to interoception and empathy), the hippocampus (memory consolidation).

You can't place an EEG electrode on the anterior cingulate. It's buried in the medial wall of the frontal lobe. So how do you train it?

LORETA (Low Resolution Electromagnetic Tomography) uses a mathematical trick. If you have enough surface electrodes, and if you know the electrical conductivity properties of the skull, cerebrospinal fluid, and brain tissue, you can run the physics backward. You can take the pattern of voltages on the scalp surface and estimate which deep brain structures most likely produced that pattern.

It's called the "inverse problem," and solving it is one of the hardest computational challenges in neuroscience. LORETA doesn't solve it perfectly (hence "low resolution" in the name), but it solves it well enough to localize activity to roughly cubic-centimeter voxels inside the brain. That's not as precise as an fMRI, but it's precise enough to distinguish the anterior cingulate from the prefrontal cortex from the insula.

The promise

LORETA neurofeedback is particularly exciting for conditions involving deep brain structures. Obsessive-compulsive tendencies, for example, often involve hyperactivity in the anterior cingulate cortex. Depression frequently shows abnormal activity in the subgenual cingulate. These structures are beyond the reach of surface neurofeedback, but LORETA can target them (at least in theory).

The limitations

LORETA requires a full 19-channel EEG cap, a quantitative EEG (QEEG) brain map to guide the protocol, specialized software, and a trained clinician to interpret the results. It is not a home-use technology. The spatial resolution, while useful, is still limited compared to fMRI. And the evidence base, while growing, consists mostly of case series and small controlled trials rather than the large-scale randomized studies that support frequency training for ADHD.

There's also a conceptual debate about whether the mathematical assumptions underlying LORETA hold up well enough to justify clinical use. The inverse problem has infinite possible solutions, and LORETA picks one by making assumptions about which solution is "smoothest." Some researchers argue these assumptions introduce too much uncertainty for clinical decision-making.

Type 4: Hemoencephalography (The One That's Not Even EEG)

Here's the "I had no idea" entry in the neurofeedback family.

Hemoencephalography, or HEG, doesn't measure brainwaves at all. It measures blood flow. Specifically, it measures the oxygenation of blood in the prefrontal cortex using infrared light, similar in principle to how a pulse oximeter on your finger measures blood oxygen.

The logic goes like this: when a brain region is active, it needs more oxygen. Blood flow increases to deliver that oxygen. By measuring prefrontal blood oxygenation in real time and rewarding increases, you're effectively training the brain to boost metabolic activity in the prefrontal cortex, the region responsible for executive function, impulse control, working memory, and emotional regulation.

There are two flavors. Near-infrared HEG (nirHEG), developed by Hershel Toomim, shines near-infrared light through the forehead and measures how much is absorbed by oxygenated versus deoxygenated hemoglobin. Passive infrared HEG (pirHEG), developed by Jeffrey Carmen, measures the thermal radiation emitted by the prefrontal cortex, which increases with metabolic activity.

Both place a sensor on the forehead. Both provide feedback about prefrontal activity. Neither involves EEG electrodes, conductive gel, or sensitivity to electrical noise.

Why it exists

HEG was developed partly out of frustration with EEG's limitations. EEG signals on the forehead are notoriously contaminated by muscle artifacts from the frontalis muscle (the one that wrinkles your forehead). Eye blinks, eye movements, and jaw clenching all generate electrical noise that can overwhelm the brain signal at frontal sites. HEG sidesteps this problem entirely because infrared light doesn't care about muscle electricity.

The evidence

The research base for HEG is smaller than for EEG-based neurofeedback. The strongest evidence is for migraines. Jeffrey Carmen published a case series showing that pirHEG reduced migraine frequency by over 50% in most participants, with effects lasting at follow-up. Several other small studies support HEG for attention deficits and executive function improvement. But we're still waiting for the kind of large randomized controlled trials that would put HEG on firmer scientific ground.

The equipment

HEG requires a specialized infrared sensor headband that sits on the forehead. These are not interchangeable with EEG devices. You can't do HEG with the Crown or any other EEG headset, and you can't do EEG neurofeedback with an HEG device. They measure fundamentally different signals.

Brainwave data, captured at 256Hz across 8 channels, processed on-device. The Crown's open SDKs let developers build brain-responsive applications.

Explore the Crown

Type 5: Infra-Low Frequency Neurofeedback (The Controversial One)

If frequency training operates in the range of 1-40 Hz and most clinical protocols live between 4 and 30 Hz, infra-low frequency (ILF) neurofeedback goes way, way below that. We're talking about training cortical oscillations under 0.1 Hz. Some ILF practitioners work with frequencies as low as 0.001 Hz.

At those frequencies, a single wave cycle takes 10 seconds to over 15 minutes to complete. These aren't the rapid oscillations you see in a typical EEG readout. They're ultra-slow fluctuations in cortical excitability that unfold on the timescale of mood, arousal, and autonomic regulation.

The theoretical basis is intriguing. These slow cortical potentials appear to modulate the brain's overall excitability threshold, essentially setting the gain knob for all the faster oscillations that ride on top of them. Training the infra-low frequencies, proponents argue, has cascading effects on the entire EEG spectrum because you're adjusting the foundation that everything else sits on.

The practice

ILF neurofeedback looks different from other types. Sessions tend to be more exploratory. The clinician adjusts the target frequency during the session based on the patient's subjective reports ("I feel more alert," "I feel calmer," "I feel spacey"). The optimal frequency is found through a titration process unique to each individual. It's more art than algorithm.

The controversy

ILF neurofeedback is perhaps the most debated type in the field. Skeptics question whether signals that slow can be reliably measured with standard EEG equipment, since most amplifiers filter out frequencies below 0.5 Hz. ILF practitioners use specialized DC-coupled amplifiers that can record these ultra-slow fluctuations, but questions remain about whether the signal reflects genuine cortical oscillations or slow-drift artifacts from the electrodes.

The evidence base is limited to case reports and small uncontrolled studies. No large randomized controlled trials have been published as of 2026. Proponents report dramatic clinical results across a wide range of conditions (ADHD, PTSD, autism, chronic pain), but the plural of anecdote is not data, and the field needs rigorous controlled research.

The equipment

ILF neurofeedback requires DC-coupled EEG amplifiers that don't filter out ultra-low frequencies. Standard EEG devices, including most consumer headsets, apply a high-pass filter (typically at 0.5 Hz) that removes exactly the signal ILF training targets. This makes ILF neurofeedback a clinic-only technology with specialized hardware requirements.

Type 6: Alpha-Theta Training (The Altered State)

Alpha-theta neurofeedback occupies a unique niche. Instead of training you to produce more of a "good" frequency during an alert waking state, it deliberately guides your brain toward the boundary between wakefulness and sleep.

Here's the setup. You sit in a comfortable chair with your eyes closed. The room is quiet and dark. EEG electrodes monitor alpha (8-13 Hz) and theta (4-8 Hz) activity, usually at the back of the head (Pz or O1/O2). Two tones play continuously: one represents alpha amplitude, the other theta amplitude. As you relax, your alpha drops and your theta rises. When theta amplitude crosses above alpha, that's the "crossover point," and it corresponds to a hypnagogic state. You're on the edge of sleep. Imagery flows freely. Memories surface. Emotional material that's normally suppressed by waking consciousness becomes accessible.

The therapeutic value of the crossover state is what makes alpha-theta distinct from other neurofeedback types. It's not about optimizing cognitive performance. It's about accessing deeply buried emotional content in a controlled, supported way. The auditory tones keep you hovering at the crossover point instead of falling fully asleep. You're conscious enough to process the material that surfaces, but relaxed enough that your defenses are down.

The evidence

Alpha-theta training has its strongest evidence in two domains: PTSD and substance use disorders. The Peniston Protocol, developed by Eugene Peniston in the late 1980s, combined alpha-theta neurofeedback with guided imagery for treating combat veterans with alcohol dependence. His original study showed dramatic reductions in relapse rates (contrasted with a control group that mostly relapsed within a year). Several subsequent studies have replicated the general finding, though methodological quality has varied.

For PTSD, alpha-theta training has shown promising results in reducing hyperarousal symptoms and improving emotional regulation. The theoretical fit is strong: PTSD involves a nervous system stuck in a hypervigilant state, and alpha-theta training directly trains the brain to access deep relaxation without dissociation.

The equipment

Alpha-theta training requires surprisingly little hardware. One or two EEG channels at posterior sites, a dark quiet room, and software that generates the two tones based on alpha and theta amplitude. The clinical challenge isn't the technology. It's the therapeutic skill required to support patients through the emotional material that surfaces during the crossover state. This is genuinely not a solo-at-home practice for people working through trauma. A trained therapist should be present.

Comparing All Six: The Big Picture

Now that you've met each type, here's how they stack up side by side.

A few patterns jump out of this table.

Evidence and accessibility are inversely correlated. The most well-supported types (frequency training, Z-score) are also the most accessible to consumers and home users. The more exotic types (LORETA, ILF) require specialized clinical equipment.

There's a tradeoff between specificity and complexity. Frequency training is conceptually simple but requires expert protocol design. Z-score automates protocol decisions but assumes "normal" is optimal. LORETA offers deep-brain specificity but demands 19 channels and heavy computation.

Not all types are mutually exclusive. Many clinicians combine approaches. A common pairing is frequency training for the first 20 sessions (to address the primary symptom) followed by alpha-theta training for 10 sessions (to process underlying emotional material). Z-score training can be overlaid on frequency training. The types are tools in a toolkit, not competing religions.

How to Think About Choosing a Type

If you're reading this because you're considering neurofeedback for yourself, here's a practical framework.

Start with what has the strongest evidence for your specific goal. If it's ADHD or attention issues, frequency training (specifically theta/beta ratio training at frontal and central sites) has the deepest evidence. If it's anxiety, both frequency training (typically SMR uptraining or alpha uptraining) and alpha-theta training have solid support. If it's migraines and you've exhausted other options, HEG is worth exploring.

Consider your access to equipment and expertise. Frequency training and Z-score neurofeedback are the two types most accessible to home users with consumer EEG devices. The Neurosity Crown's 8 channels and open SDKs make it possible to build and run both types of protocols without a clinical setup. LORETA requires a 19-channel cap and a QEEG brain map interpreted by a specialist. ILF requires DC-coupled amplifiers. HEG requires infrared hardware.

Be skeptical of practitioners who claim one type treats everything. Each type has a specific mechanism and a specific evidence base. A clinician who insists that LORETA neurofeedback (or ILF, or any single type) is the answer to every neurological condition is either oversimplifying or selling you something.

Pay attention to hardware quality. Regardless of which type you choose, the quality of the measurement determines the quality of the feedback. Neurofeedback only works if the signal is reliable enough for the brain to detect a consistent contingency between its activity and the reward. Noisy data means inconsistent feedback, and inconsistent feedback means broken operant conditioning. This is why sample rate, channel count, and signal processing matter as much as the protocol itself.

The Part That Should Rewire How You Think About Your Brain

Here's what's genuinely remarkable about the neurofeedback landscape when you step back and look at the whole thing.

Six fundamentally different techniques. Different signals, different sensors, different mechanisms, different clinical targets. And yet they all converge on the same discovery: the brain can change itself when given information about itself.

Train a specific frequency? The brain adjusts. Normalize multiple metrics against a database? The brain adapts. Estimate deep-source activity and feed it back? The brain responds. Measure blood flow with infrared? The brain increases perfusion. Slow the oscillations down to one cycle per minute? The brain apparently modulates. Guide the brain to the edge of sleep? It accesses emotional material it normally walls off.

The brain is not passive hardware. It's a self-modifying system that responds to any feedback channel you open. The types of neurofeedback don't disagree about this. They're all different experiments that keep landing on the same conclusion.

And the tools to run these experiments keep getting more accessible. An 8-channel EEG device with 256 Hz sampling and on-device processing can sit on your desk right now. Open SDKs mean you don't need a clinical license to explore what your brain does when you close the feedback loop. The gap between "what researchers can study" and "what curious people can try" has never been narrower.

The neurofeedback field has spent decades arguing about which type is best. But maybe the more interesting question is what happens when everyone has the hardware to find out for themselves.