How it works
How an intraoral scanner actually works
A scanner never sees a whole arch. Its window is about 18 mm, which is two or three teeth. Almost everything that matters when you are buying one follows from that single fact.
Updated 9 September 2026 · 7 min read
It is a chain, not a photograph
Because the window is small, the scanner captures hundreds of little 3D frames and aligns each new one onto the growing model in real time, matching them by the part they have in common. That alignment step is called registration, and it is the whole trick.
A single tooth is one or two frames, so a bench figure of 5 to 10 microns is a fair description of it. A full arch is a chain of hundreds of alignments, each contributing a small rotational error that the next one inherits. That is why clinical full-arch trueness sits between 20 and 60 microns for every scanner on the market, and why the error is always worst at the back.
How the depth is measured
Structured light
Project a known pattern of stripes or dots onto the teeth. Cameras sitting at a known angle see that pattern bent by the surface, and trigonometry turns the bend into a depth value for every pixel. Most scanners sold today work this way, including the ones whose spec sheets mention coded structured light.
Confocal
Sweep the focal plane through the tooth instead. Any given point is sharp at exactly one depth, so the scanner finds the depth where sharpness peaks and reads the height from that. The classic iTero line was built on this approach.
Why a tooth is a hostile target
Scanning an opaque object on a bench is a solved problem. A mouth is not a bench.
- Enamel is translucent. Light sinks in and scatters before it comes back, so the surface reads deeper than it actually is.
- Saliva is mirror-like, and it moves. Specular highlights confuse a system that assumes light scatters off the surface it hits.
- Metal restorations blow out the sensor, and gingiva deforms under the tip, so the thing you are measuring changes shape while you measure it.
That combination is why scanners once needed a dusting of titanium dioxide powder, and why the tip is actively heated today: a cold lens in a warm wet mouth fogs instantly. All of it has to happen at 30 to 70 three-dimensional frames a second, because the patient is breathing and the tongue is moving. That is what the discrete GPU and 32 GB of RAM in the spec sheet are actually for.
The worst case is an arch with no teeth
Registration works by matching distinctive geometry between one frame and the next. A bare ridge has almost none: it is smooth, repetitive and soft. The algorithm has nothing to lock onto, so drift grows far faster than it does along a dentate arch. This is why reviewers single out particular scanners for edentulous performance rather than treating it as a given, and why it is worth asking about specifically if you do denture or full-arch implant work.
Where the money actually goes
Measuring one frame is close to a commodity now. A mature contract manufacturing supply chain sells competent sensor and projector modules to anyone, which is why several scanners on the market share a spec sheet under different brand names.
What cannot be bought in is the registration algorithm. It has an enormous fixed cost and no marginal cost, which is the classic shape of software. But cost and price are different things, and the difference is where your negotiating room lives.
Registration software
Real costKeeping drift low at the back of the arch, on wet translucent enamel, at sixty frames a second. Hard, ongoing, and the one thing that cannot be bought in.
Regulatory clearance
Real costFDA, EU MDR, and each national regulator, per market and forever. Small badges can only afford a few markets, which is why some scanners never leave their home region.
A small market
Real costTens of thousands of units a year worldwide. A phone camera module amortises its research over billions, which is why it costs a few dollars and this does not.
Dealer margin
MarginDental distribution commonly takes 30 to 40 percent. This is the single biggest reason list price is fiction and two quotes are worth more than any spec sheet.
Ecosystem capture
MarginA compulsory monthly fee is rarely recovering a cost. Where the manufacturer also sells the aligners or the mill, the scanner is priced as the toll booth rather than the product.
The last two are why list price is fiction, and why getting a second dealer quote is worth more than any specification you will read.
What this means when you are buying
A budget scanner and a premium one can produce near-identical scans of a single crown prep, because that is one or two frames and the optics behind them are largely the same parts. They diverge across a full arch, because that is hundreds of alignments and the algorithm doing them is proprietary.
So the question is not which scanner is more accurate in the abstract. It is whether the work you actually do lives at the end of a long chain of frames. Orthodontic records and aligner cases are full arch, but aligner laboratories work to tolerances where the difference rarely changes the appliance. Deep margins and full-arch implant cases are where the extra money starts earning.
This page is part of our complete guide to intraoral scanners. If you are choosing between two right now, the scanner picker will narrow the field, and the overlay test on that page will settle it with a measurement rather than an argument.
Questions
Why is full-arch accuracy always worse than the brochure figure?
Because the brochure figure describes one frame and a full arch is hundreds of them. A scanner sees about 18 mm at a time, so it captures a long chain of small 3D frames and aligns each onto the previous one by their overlap. Every alignment adds a tiny rotational error, and those errors compound along the chain. A single tooth is one or two frames, which is why bench figures of 5 to 10 microns are honest. A full arch is a chain, which is why real clinical trueness lands between 20 and 60 microns regardless of whose logo is on the handpiece.
How does a scanner measure depth at all?
Two methods dominate. Structured light projects a known pattern of stripes or dots onto the teeth; cameras sitting at a known angle see that pattern bent by the surface, and trigonometry turns the bend into a depth value for every pixel. Confocal scanning instead sweeps the focal plane through the tooth: any given point is only sharp at one depth, so the scanner finds the depth at which sharpness peaks. Most scanners sold today use structured light.
Why do teeth make this harder than scanning an ordinary object?
Enamel is translucent, so light sinks in and scatters before returning, which makes the surface read deeper than it is. Saliva is mirror-like and moves. Metal restorations blow out the sensor. Gingiva deforms under the tip. That combination is why scanners once needed titanium dioxide powder, and why the tip is actively heated: a cold lens in a warm wet mouth fogs immediately.
Why are edentulous arches the worst case?
Because the alignment algorithm matches each new frame to the last using distinctive geometry, and a bare ridge has almost none. Smooth, repetitive tissue gives the algorithm nothing to lock onto, so drift grows faster than it would along a dentate arch. This is why reviewers single out particular scanners for edentulous performance, and why it is worth asking about specifically if you do denture or full-arch implant work.
If the optics are commodity, what am I paying for?
Registration software, regulatory coverage, and margin. The sensor and projector modules can be bought from a mature contract-manufacturing supply chain, which is why several scanners sold under different brands share a spec sheet. What cannot be bought in is the algorithm that keeps drift low at the second molar. Add the cost of clearing FDA, EU MDR and each national regulator, then the 30 to 40 percent that dental distribution commonly takes, and you have most of the price.
Your scans are only worth what you can find later.
Kythro keeps scans, plans and photos on the patient record. Start a free 30 day trial.