A small spread does not mean the measurements are close to an independent reference.
Centered, but variable
A mean close to the reference can hide a large spread between runs. Check both.
Conceptual positions, not scanner measurements. Target center represents the reference.
Accuracy asks whether the captured dimension agrees with an independent reference. Resolution asks whether a feature can be resolved. A smaller point spacing does not prove a smaller dimensional error.
For a long part, look for a length-dependent claim as well as a point figure. Artec Leo's published 0.1 mm + 0.3 mm per meter gives 0.4 mm at 1 m. That calculation illustrates the specification; it does not predict the result on your object.
Your next step: choose one critical feature, measure it independently, then compare dimensions from fresh captures. Use the calculator below, and open the supporting definitions and specification tables when you need them.
The smallest number on a product page is rarely the one that decides your purchase. Before comparing scanners, find out what each number describes, how it was measured and whether those conditions look anything like your part. A point-distance setting, for example, says how densely the surface is sampled, not how far the result is from the real object.
This guide explains the terms, shows how real manufacturer claims differ, and ends with a sheet for checking accuracy on your own part during a demo. The source dates are listed with the claim table and worked length formulas.
Calculate your demo differences
The worksheet below keeps two questions separate: how far your entered dimensions sit from the reference, and how far they sit from each other. Enter measurements of the same feature from separate captures, not repeated clicks on one mesh.
The mean signed difference is the average of measured value minus reference value. The observed range is the largest entered dimension minus the smallest. For example, 100.08, 100.09 and 100.10 mm against a 100 mm reference have a mean difference of +0.09 mm and a range of 0.02 mm. Close agreement between runs does not remove the offset.
Use the calculation to compare your measurements with a reference. A few runs do not establish a stable bias or repeatability specification. NIST guidance explains measurement uncertainty, which also depends on the reference, setup and other error sources. Keep those conditions in the demo sheet before deciding whether the result meets your tolerance.
Read actual claims side by side
Compare the exact maker claims and their limits
The examples below show why accuracy, resolution and point spacing need separate labels. Check the quantity and test conditions behind each manufacturer figure.
Model
Official claim
How to read it
EINSTAR VEGA
Point distance 0.05-3 mm in HD mode, 0.5-10 mm in Fast mode
Point spacing by mode. EINSTAR does not publish an accuracy figure for VEGA
Revopoint MIRACO Plus
Single-frame accuracy up to 0.04 mm; photogrammetric length accuracy 0.02 mm + 0.05 mm per meter
A single-frame figure and a separate length figure for its photogrammetry kit
Peel 3
Accuracy up to 0.050 mm; volumetric accuracy 0.050 mm + 0.100 mm per meter
A best-case figure plus a length-dependent one
Artec Leo
3D point accuracy up to 0.1 mm; 3D resolution up to 0.2 mm; accuracy over distance up to 0.1 mm + 0.3 mm per meter
Three different statements, not three ways of saying 0.1 mm
Artec Point II
3D point accuracy up to 0.02 mm; volumetric accuracy up to 0.015 mm + 0.035 mm per meter
Manufacturer references VDI/VDE 2634 and JJF 1951 testing. ISO/IEC 17025 concerns laboratory competence; request the applicable calibration record and uncertainty
HandySCAN BLACK Elite
Accuracy 0.025 mm; volumetric accuracy 0.020 mm + 0.040 mm per meter
Applies to the BLACK Elite; do not borrow figures from other HandySCAN editions
To see what a volumetric formula means, put a length into it. For HandySCAN BLACK Elite, 0.5 m gives 0.020 + 0.020 = 0.040 mm, and 1 m gives 0.060 mm. That is arithmetic on the published claim, not a prediction for your part.
The VEGA guide discusses what a missing accuracy figure means in practice, and the VEGA vs MIRACO Plus comparison keeps both scanners in the same test instead of declaring a winner from unlike numbers.
Read a length-dependent claim
Use the named measuring quantity and the units printed beside the formula. A scanner's distance from the surface, the object's overall envelope and the span you measure are different things. A smaller calculated value from another method does not by itself identify the better scanner.
Worked formulas and your own span
These are calculations on two named manufacturer specifications. They do not predict the result on your part or make unlike test methods comparable.
Both source examples at 0.5, 1 and 3 meters
Artec Leo
3D accuracy over distance (up to)
Formula: 0.1 mm + 0.3 mm per meter.
Length in meters in the distance specification. Confirm the applicable measuring span and test conditions; this is separate from the point-accuracy figure.
L is the maximum measuring distance in meters as defined by Revopoint. This formula belongs to the Photogrammetric Metrology Kit workflow; it is separate from single-frame accuracy.
Read accuracy, resolution, precision and repeatability definitions
Manufacturers do not all define these terms the same way. When a number will influence your decision, ask for the test method behind it.
Term
What it describes
What it does not tell you
Accuracy
How close a measurement is to a reference, under the conditions the manufacturer states
That every dimension of every object will meet the headline figure
Volumetric accuracy
How error grows over a measuring volume or length, often written as a fixed part plus an amount per meter
That it can be compared with a single-point accuracy figure
Resolution
How finely the scanner or software resolves detail, as the manufacturer defines it
That fine detail is dimensionally correct
Point distance
The spacing of points in a given mode
That this spacing is the measurement error
Precision
How closely repeated results agree with each other
That those tightly grouped results agree with the reference
Repeatability
How closely results agree when the setup, operator and conditions stay the same
How results change with a different operator, day or setup
A scanner can repeat the same offset every time and still miss the reference. A dense mesh can preserve noise as faithfully as real detail. Keep the reference check and the repeat-scan check as two separate pieces of evidence.
Can a finer point distance rescue a missed feature?
Reducing a sampling interval can retain more detail only where the capture actually contains usable surface information. It cannot restore a hidden wall, correct a common dimensional offset or turn a filled hole into observed geometry. Exported triangle density can also change during processing.
Inspect the sampling example and its limits
Shift the grid, keep the spacing
Move the blue sample positions across the shaded 0.040 mm interval. Their spacing stays 0.050 mm.
Arrow keys: 0.001 mm. Home: 0.000 mm. End: 0.049 mm.
First sample: 0.000 mm
Horizontal position (mm)
0 samples strictly inside; none on a boundary.
Sample positions: 0.000, 0.050, 0.100 mm.
Compare: first sample at 0.025 mm
Horizontal position (mm)
1 sample strictly inside; none on a boundary.
Sample positions: 0.025, 0.075, 0.125 mm.
Illustrative numbers, not scanner measurements. A sample inside the interval alone does not resolve both edges or the width.
This simple example shows why point spacing alone is not enough. Assume a narrow recess occupies positions 0.005 to 0.045 mm on a profile, a width of 0.04 mm. Both sets below have 0.05 mm spacing:
Sample positions near the recess, mm
A sample inside that interval?
What the example establishes
0.000, 0.050, 0.100
No
This placement misses the interval
0.025, 0.075, 0.125
Yes, at 0.025
A point falls inside; both edges and the width still need to be resolved
Example sampling arithmetic, not a scanner measurement. Optical access, material, noise, alignment and processing also affect whether a real feature survives.
Ask the vendor to preserve the named feature in raw geometry, state the mode and sampling setting, then repeat the required measurement after cleanup and reconstruction. Use the feature-based reverse-engineering brief. The calculator compares entered dimensions with a reference; it does not determine optical resolution.
Record your measurement checks
Keep the setup and independent reference beside your repeated captures. These blank templates contain no results, preset limits or formulas.
Read the complete demo sequence and context checklist
Agree on pass and fail limits before the vendor sees any result. If your reference measurement is not good enough to support the limit you want, choose a better reference method first.
Record
What to fill in before the demo
Job
Part, material, overall size and required deliverable
Critical dimensions
Named features, nominal values and pass/fail limits
Reference
Instrument or artifact, its calibration status and uncertainty
Setup
Exact scanner, mode, software, calibration, lighting and surface preparation
Processing
Alignment, smoothing, hole filling and mesh reduction settings
Results
Reference value, scanned value and signed difference for each dimension
Repeats
Independent captures with the same setup, each recorded separately
Decision
Pass, fail or insufficient evidence, with the reason
A practical sequence
Choose a representative part with one short feature and one long span. Mark which surfaces or edges define each measurement.
Measure the reference independently, for example with calipers, a height gauge or a CMM report. Do not use the scan as its own reference.
Capture and export with the workflow you are being quoted. Keep both unprocessed and processed files.
Measure the agreed features with the agreed alignment. Note missing geometry instead of letting it be filled in.
Repeat the capture from a fresh start. Three runs make a useful demo, but they are not a statistical validation.
If a result fails, find out why before changing the limit. A second operator can be tested as a separate condition, not as a quiet replacement for the failed run.
Match your acceptance question to the evidence needed
Question
Check to run
What remains unresolved
Was the feature captured?
Inspect unedited geometry and optical access
A complete-looking surface is not an independent dimensional result
Is the dimension close to a reference?
Same feature definition, stated alignment and independent reference
Reference uncertainty and the full measurement process
Do independent captures agree?
Fresh capture for each run with the same documented setup
Agreement alone can retain a common offset
Does the long span agree?
Check a suitable span/reference, not only a small local detail
Conditions and length-dependent specification
Did processing change the result?
Compare retained raw/full geometry with the processed delivery
Whether the chosen reconstruction is acceptable design intent
Does the reconstructed part work?
Receiving edit, declared datums and fit/inspection checks
Physical fit, manufacture and functional use
Define the decision rule before a result appears. If the reference's uncertainty is not suitable for your tolerance, the outcome is insufficient evidence. Do not use a caliper reading as a universal reference for every geometry or tolerance.
Point distance, triangle count and a minimum object cube answer different questions from a minimum functional feature. For a bore or recess, visibility and processing may decide whether a nominally dense capture retains it. Ask the seller to name the mode and preserve the feature before hole filling.
Compare setup changes openly
Keep independent captures under one condition together. A second operator, another day, recalibration, coating or another processing setting is a new condition. Report it separately. Preserve failed captures as well as successful ones.
Use the calculator for signed differences and observed spread. Its arithmetic does not estimate a complete uncertainty budget or certify repeatability from a few runs. The example datasets illustrate a common offset versus dispersed values.
Ask what the standard and certificate cover
Read the standards and calibration-record checks
VDI/VDE 2634 is a German guideline series for evaluating optical 3D measuring systems. A reference to it only means something when you know which part and procedure was used. Creaform, for example, states that its HandySCAN BLACK acceptance test is based on Part 3. That describes the manufacturer's test of its system, not every object you will measure later.
NIST research on structured-light scanners discusses these tests and the error sources they can and cannot detect. Treat a standard-based result as evidence with a defined scope.
ISO/IEC 17025 covers the competence of testing and calibration laboratories ↗. It is not a certificate for a scanner model and not a promise that your next scan meets your tolerance. Ask for the calibration record itself: date, serial number, stated uncertainty and the laboratory's scope, and check whether the quote includes recalibration.
Describe the measured difference precisely
In its measurement handbook ↗, NIST treats accuracy as qualitative agreement with a reference and bias as a quantitative difference involving averages. The calculator reports the mean signed difference and observed range of the values you enter. A few demo captures are not a validated bias or repeatability specification.
NIST's uncertainty framework ↗ also ties uncertainty to the particular result and measurement configuration. Record the reference, operator, alignment, environment and processing with the result.
Inspect community scan examples
OpenScanBenchy collects community captures of a small resin figurine. These examples use different prints, operators, preparation and processing. Read the setup notes, then compare your own part under the same conditions.
View 5 scanner examples and their setup notes
Creality Otter
Contributor: OpenScan_Thomas
Reported setup
Object tracking, highest meshing quality and no smoothing are reported.
Before comparing
A meshing-quality setting is not a dimensional test.
These outside contributions can help you spot missing features and ask better demo questions. For editable mesh inspection, use the scan-to-CAD file checks. For a dimensional decision, keep the independent reference and repeated-capture checks above.
Frequently asked questions
How accurate are 3D scanners?
Manufacturer claims for the handheld scanners on this site range from about 0.02 mm to 0.1 mm, and some models publish no accuracy figure at all. Those figures come from the manufacturers' own test conditions. On your part, accuracy also depends on surface, size, alignment, processing and the operator, so check it with a reference measurement in a demo.
What is the difference between accuracy and resolution?
Accuracy is how close the result is to the true dimension. Resolution is how much detail the scanner can capture. A scanner can capture fine detail and still be offset from the true size, which is why the two must be checked separately.