What Is LORETA Neurofeedback?

The 2D Problem Nobody Noticed for 40 Years

Traditional neurofeedback has a secret that practitioners don't always talk about upfront. It's a flatland technique.

When you stick an electrode at position Cz on someone's scalp (that's the spot right at the top of your head) and train the brain to produce more of a particular frequency there, you're making an assumption. You're assuming that the signal at Cz represents activity happening in the cortex directly below Cz. And for 40 years, that assumption was considered good enough.

It isn't.

Here's why. The electrical signal your scalp electrode picks up doesn't come from a single spot in the brain beneath it. Thanks to a phenomenon called volume conduction, that signal is a blurred mixture of activity from many sources, some nearby, some surprisingly far away. The skull acts like a bad lens, smearing and distorting the electrical patterns as they pass through. By the time neural activity reaches the surface, the spatial information is garbled.

This means when you train "beta at Cz," you're not necessarily training the brain region under Cz. You're training whatever blend of sources happens to show up at that electrode location. It might be motor cortex activity from directly below. It might include contributions from the anterior cingulate, a structure buried deep in the medial wall of the brain. It could even pick up smeared signals from the parietal lobe.

You're training a shadow on the wall. And you have no idea what's casting it.

In the early 1990s, a Zurich-based neuroscientist named Roberto Pascual-Marqui looked at this problem and had a simple but profound thought: what if, instead of training the shadow, we could figure out what's making it?

The Math That Looks Through Your Skull

To understand LORETA neurofeedback, you first need to understand the idea it's built on: source localization.

Source localization is the attempt to work backward from what EEG measures on the scalp to figure out where inside the brain those signals originate. This is known as the inverse problem, and it's one of the genuinely hard problems in all of physics. Hermann von Helmholtz proved in the 1850s that the inverse problem has no unique solution. There are infinitely many possible arrangements of sources inside the brain that could produce the exact same pattern on the scalp.

Infinitely many. Not "a lot." Not "thousands." Literally infinite.

So how does anyone solve it? By adding assumptions. Every source localization method constrains the infinite solution space by making claims about how the brain probably works. Different methods make different assumptions, and those assumptions determine what the method is good at.

Pascual-Marqui's insight was elegant. He divided the brain volume into thousands of tiny cubes called voxels (about 6,239 of them in the original implementation, each 7mm on a side). He placed a potential electrical source at every voxel. Then he asked: given the signals I see on the scalp, what's the smoothest possible distribution of activity across all those voxels that explains the data?

That smoothness constraint is the key. LORETA assumes that if a neuron in one spot is active, its neighbors are probably active too. This is biologically reasonable. Neurons that are physically close to each other tend to be wired together into functional groups that fire in correlated patterns. A single isolated voxel blazing away while everything around it sits silent would be neurologically bizarre.

By enforcing smoothness, the infinite solution space collapses to a single unique answer. And that answer, while not perfect, turns out to be remarkably useful.

The name captures the tradeoff honestly: Low Resolution Electromagnetic Tomography. LORETA tells you the general neighborhood of brain activity, not the exact address. But knowing the neighborhood turns out to be a massive upgrade over knowing nothing at all.

From LORETA to sLORETA to eLORETA: The Three Generations

Pascual-Marqui didn't stop with his 1994 paper. He kept refining the math, and the three versions of LORETA represent a genuine progression in accuracy.

LORETA (1994) works, but it has a localization bias. It tends to place estimated sources deeper and more central than they actually are. This is a known mathematical artifact of the smoothness constraint. If you're looking at genuine activity in the dorsolateral prefrontal cortex, LORETA might estimate it as coming from slightly deeper and more medial.

sLORETA (2002) fixed this by standardizing the current density estimates. Think of it like grading on a curve, but for brain voxels. Instead of just reporting the estimated current at each voxel, sLORETA divides each estimate by its own variance, which corrects for the fact that some voxels are inherently easier to estimate than others. The result, confirmed by independent validation studies, is zero localization error for single point sources. If one spot in the brain is generating a signal, sLORETA will find it with no systematic bias.

eLORETA (2007) went further. It provides zero localization error not just for single sources but for any source configuration, even when multiple regions are active simultaneously. The math behind it is more complex (it uses a weighted minimum norm approach with exact error correction), but the practical result is straightforward: eLORETA is the most accurate of the three and is currently considered the gold standard.

Here's the part that surprises most people: sLORETA and eLORETA are free software. The KEY Institute at the University of Zurich, where Pascual-Marqui developed them, distributes the software at no cost for research and clinical use. This accessibility is a big reason why LORETA-based methods have spread as widely as they have.

So What Is LORETA Neurofeedback, Actually?

Now we can put the pieces together.

Traditional neurofeedback reads your EEG, picks a frequency band at a scalp electrode, and trains you to change the power in that band. The feedback loop goes: scalp signal to analysis to reward.

LORETA neurofeedback adds a critical middle step. It reads your EEG from all 19 (or more) channels simultaneously, runs the sLORETA or eLORETA inverse model in real time, estimates the current density at a specific region deep inside the brain, and trains you to change the activity at that estimated source.

The feedback loop becomes: scalp signals (all channels) to source localization to activity estimate at target region to reward.

This is a fundamentally different thing. Instead of training "more beta at electrode Fz," a LORETA neurofeedback clinician can say "I want to train increased beta current density in the anterior cingulate cortex, Brodmann area 24." Instead of working with the blurred shadow on the wall, they're working with a computed estimate of the thing casting the shadow.

Why Clinicians Got Excited About Going 3D

The appeal of LORETA neurofeedback makes intuitive sense once you see the anatomy. Many of the brain structures most involved in clinical conditions sit deep in the brain, tucked away from the scalp where traditional neurofeedback operates.

The anterior cingulate cortex (ACC) is buried in the medial wall between the two hemispheres. It plays a central role in error monitoring, emotional regulation, pain processing, and attention allocation. Dysfunction in the ACC has been implicated in ADHD brain patterns, OCD, depression, chronic pain, and anxiety. You can't place an electrode directly over the ACC because there's no scalp location that sits on top of it. It's deep, medial, and its signals are heavily mixed with other sources by the time they reach the surface.

The insular cortex sits folded inside the lateral sulcus, hidden between the temporal and frontal lobes. It's involved in interoception (awareness of your body's internal states), emotional processing, and the experience of disgust, pain, and empathy. Insular dysfunction shows up in addiction, eating disorders, anxiety disorders, and chronic pain syndromes.

Brodmann area 25, the subgenual cingulate, has been called "the sadness center" of the brain. Deep brain stimulation targeting this region has shown remarkable results for treatment-resistant depression. LORETA neurofeedback offers a non-invasive way to train activity in this area, though the evidence is much more preliminary.

None of these structures can be specifically targeted by placing an electrode on the scalp and training whatever shows up there. LORETA neurofeedback's promise is that it can reach them computationally, even though it can't reach them physically.

What the Evidence Actually Says

Let's be careful here. LORETA neurofeedback is a younger technique than traditional surface neurofeedback, and the evidence base reflects that. There are promising results, genuine clinical enthusiasm, and a growing body of published studies. But the large-scale randomized controlled trials that would make the case definitive haven't all been done yet.

Here's what we know.

Case Studies and Small Clinical Trials

The bulk of the published LORETA neurofeedback literature consists of case studies and small controlled trials. A 2012 study by Congedo and colleagues demonstrated that LORETA neurofeedback targeting the anterior cingulate cortex produced significant improvements in attention and executive function in adults with ADHD. The sample was small but the results were encouraging.

A case series by Cannon and colleagues published in Journal of Neurotherapy reported significant improvements in anxiety and depression symptoms using sLORETA neurofeedback targeting the anterior cingulate and insular cortex. Patients showed both symptom improvement and corresponding changes in their LORETA brain maps, suggesting the training was actually modifying activity in the targeted regions.

For OCD, several case reports have documented improvements following sLORETA neurofeedback targeting excessive activity in the anterior cingulate and orbitofrontal cortex, regions consistently implicated in the obsessive-compulsive circuit.

The Comparison Problem

Here's the honest complication. It's hard to know whether LORETA neurofeedback works because it targets deep sources specifically, or whether it works for the same reasons traditional neurofeedback works and the source localization is just a more sophisticated way of doing roughly the same thing.

Think about it. When you train "beta in the anterior cingulate via sLORETA," the feedback signal the client receives is still derived from all 19 scalp channels. The brain still has to figure out how to change its activity to get the reward. The question is whether the LORETA-based feedback provides a more specific training signal that leads to more targeted brain changes, or whether the brain would make the same adjustments either way.

Some studies have compared LORETA neurofeedback directly to traditional surface neurofeedback for the same conditions. The results are mixed. Some show advantages for the LORETA approach, particularly for conditions involving deep structures. Others show equivalent outcomes. No large meta-analysis has settled the question definitively.

The Validation Studies

There is, however, good evidence that LORETA neurofeedback actually does what it claims to do mechanistically. Studies using simultaneous fMRI and EEG have confirmed that when subjects perform sLORETA neurofeedback targeting a specific brain region, the fMRI signal in that region changes accordingly. This is important because it means the source localization isn't just a mathematical abstraction. The feedback is reaching the intended target, at least regarding the brain's hemodynamic response.

A 2015 study by Vitello and colleagues used simultaneous EEG-fMRI to show that sLORETA neurofeedback targeting the anterior cingulate produced BOLD signal changes localized to the ACC. The training effect was specific to the target region, not a broad, nonspecific change in brain activity.

The Crown captures brainwave data at 256Hz across 8 channels. All processing happens on-device. Build with JavaScript or Python SDKs.

Explore the Crown

The Equipment Question: What You Actually Need

LORETA neurofeedback has a higher barrier to entry than traditional neurofeedback, and it's worth being specific about what that means.

The 19-channel minimum is not arbitrary. It comes directly from the mathematics of the inverse problem. Each EEG channel provides one spatial data point. LORETA's source localization estimates current density at thousands of brain voxels from those data points. With fewer channels, there simply isn't enough spatial information to constrain the inverse model meaningfully.

Here's a useful way to think about it. Imagine you're trying to reconstruct a photograph from a handful of pixels. With 19 pixels, you can get a rough but recognizable image. With 64 pixels, you can make out real details. With 8 pixels, you're guessing. The same principle applies to LORETA: each channel is a pixel, and you need enough of them for the reconstructed image to mean something.

This is why consumer EEG devices with 4 to 8 channels, including the Neurosity Crown, cannot perform LORETA neurofeedback. The Crown's 8 channels are strategically placed across both hemispheres and all four cortical lobes, making them excellent for surface-level neurofeedback, focus/calm monitoring, and BCI applications. But 8 spatial data points don't give the inverse model enough information to estimate the activity of a specific deep brain structure. That's not an engineering limitation. It's a mathematical one.

The "I Had No Idea" Part: Your Brain Is Already Doing What LORETA Trains

Here's the thing about LORETA neurofeedback that doesn't get discussed enough, and it genuinely changed how I think about the whole field.

The assumption behind all neurofeedback, traditional and LORETA alike, is that you're teaching the brain something new. You're training it to produce patterns it doesn't currently produce. And that's partially true.

But the deeper truth is that your brain already produces every pattern LORETA neurofeedback trains. Every single one. The anterior cingulate is already generating beta rhythms. Your insular cortex is already producing alpha. The default mode network is already cycling between activation and deactivation.

What neurofeedback actually trains isn't the existence of these patterns. It's their regulation. The timing, the amplitude, the flexibility to shift between states when the situation demands it.

Consider the anterior cingulate cortex. In a healthy brain, the ACC ramps up its activity when you make an error, when you need to override an automatic response, when conflicting information requires you to pay closer attention. Then it settles back down when the demand passes. It's not always on or always off. It modulates dynamically.

In ADHD, the ACC often shows blunted error-related activity. It doesn't ramp up enough when it should. In OCD, it shows the opposite problem: it fires too much, stays active too long, generating the sensation that something is wrong even when nothing is. In both cases, the structure exists and it works. It just doesn't regulate itself well.

LORETA neurofeedback, at its best, is teaching that regulation. Not creating new activity, but helping the brain get better at the activity it's already doing. It's the difference between teaching someone to sing versus teaching someone who can already sing to stay on pitch.

This reframing matters because it takes LORETA neurofeedback out of the realm of "zapping your brain into a new state" and places it where it belongs: alongside other forms of learning. Your brain is a learning machine. Given the right feedback about what it's doing, it can learn to do it better. LORETA just provides more specific feedback about a more specific thing.

LORETA vs. Traditional Neurofeedback: When to Choose What

This isn't an either/or situation. Many clinicians use both approaches, choosing the tool based on what the brain map reveals.

The practical question most people arrive at is: should I seek out LORETA neurofeedback instead of traditional neurofeedback?

For most people exploring neurofeedback for the first time, traditional surface neurofeedback with a well-established protocol is the more practical starting point. The evidence is stronger, the cost is lower, the barrier to entry is smaller, and for conditions like ADHD and anxiety, the outcomes are well-documented.

LORETA neurofeedback becomes most compelling when three conditions are met: a QEEG brain map identifies abnormal activity in a specific deep structure, the condition has a plausible connection to that structure's function, and traditional neurofeedback hasn't produced adequate results. In that scenario, the added precision of LORETA targeting may justify the added complexity and cost.

The Consumer EEG Reality

Let's be direct about where consumer brain-computer interfaces fit into this picture.

The Neurosity Crown has 8 EEG channels at positions CP3, C3, F5, PO3, PO4, F6, C4, and CP4. That's the highest channel count among consumer BCIs, and those positions are strategically distributed across frontal, central, centroparietal, and parieto-occipital regions. For surface-level neurofeedback, focus and calm monitoring, brainwave-driven applications, and brainwave monitoring, 8 channels provide genuinely excellent coverage.

But 8 channels cannot do LORETA neurofeedback. That's not a knock on the Crown. It's a statement about what the inverse problem requires mathematically. LORETA needs spatial information from across the full scalp to estimate deep sources, and 19 channels is the minimum that makes the math work. This is the same reason you can't create a high-resolution photograph from 8 pixels, no matter how good the camera that captured those 8 pixels is.

What 8 channels can do is powerful in its own right. The Crown computes real-time FFT, power spectral density, and frequency band power across both hemispheres. Its open JavaScript and Python SDKs let developers build custom neurofeedback protocols, brain-aware applications, and research tools. The focus and calm scoring algorithms provide immediate cognitive state feedback without any external software. And for the growing body of neurofeedback approaches that work at the surface level (theta/beta training, SMR training, alpha enhancement), 8 well-placed channels are more than sufficient.

The honest picture is that LORETA neurofeedback and consumer EEG serve different segments of the same spectrum. LORETA is the specialist tool for targeted deep-source training in a clinical setting. Consumer EEG is the accessible tool that brings brain monitoring and surface neurofeedback to anyone with a desk and curiosity about their own mind.

Where LORETA Neurofeedback Is Heading

Three developments are converging to make LORETA neurofeedback more accessible and more precise.

Better inverse models. eLORETA already provides zero localization error under ideal mathematical conditions, but the field is moving toward individual head models derived from MRI scans, which dramatically improve source estimation accuracy. As MRI becomes more accessible and head modeling software becomes more automated, the gap between population-average models and individual anatomy is closing.

Higher-density affordable EEG. The cost of multichannel EEG is dropping. Systems that cost $50,000 a decade ago can now be matched for $5,000 to $10,000. As this trend continues, the 19-channel barrier to LORETA neurofeedback becomes less of a financial obstacle, though it remains a practical one regarding setup time and expertise.

Machine learning integration. Researchers are beginning to combine LORETA source estimates with machine learning classifiers that can identify complex patterns of deep-source dysfunction more reliably than human interpretation alone. This could make LORETA neurofeedback protocols more standardized and easier for clinicians to implement.

The long-term trajectory is clear: neurofeedback is becoming more spatially specific, more personalized, and more grounded in individual brain anatomy. LORETA was the first major step in that direction. It almost certainly won't be the last.

The Bigger Picture

Step back for a moment and consider what LORETA neurofeedback actually represents.

For most of human history, the brain was a black box. We couldn't see inside it. We couldn't measure its activity. We certainly couldn't train specific internal structures to behave differently.

Then EEG came along, and suddenly we could listen to the brain's electrical activity from outside the skull. Neurofeedback showed that the brain could learn from hearing its own activity. And LORETA added the ability to listen not just to the surface noise, but to specific voices inside the box.

We're still listening through walls. The resolution is imperfect. The math requires assumptions. But the trajectory points in one direction: more precision, more accessibility, more ability for you to understand and influence the most complex object in the known universe, the one sitting between your ears.

Whether that future runs through 19-channel clinical caps doing sLORETA or through the next generation of consumer BCIs doing something we haven't invented yet, the core idea remains the same. Your brain is always talking. The question is how precisely we can learn to listen.